Pairs trading is a market-neutral strategy that bets on the convergence of two historically correlated assets. It relies on statistics rather than predicting market direction.
| Component | Description | Why It Matters |
|---|---|---|
| Pair Selection | Finding two assets with historical correlation | Bad pairs → failed trades |
| Spread Calculation | Measuring price gap between the pair | Defines entry and exit points |
| Mean Reversion | Assumption that spread returns to average | Core profit mechanism |
| Hedge Ratio | Determining position sizes for each leg | Ensures dollar neutrality |
| Exit Trigger | Signal to close when spread normalizes | Locks in profit or cuts loss |
The strategy gained fame at Morgan Stanley in the 1980s. Teams led by Nunzio Tartaglia used early computers to find price relationships.
Pepsi and Coca-Cola often move together. If Pepsi jumps 5% while Coke stays flat, a pairs trader sells Pepsi and buys Coke. They bet the gap closes.
This is not about liking Pepsi or Coke. It is about numbers.
Success depends on statistical relationships, not company fundamentals or news.
If the math breaks, the trade breaks with it.
Finding good pairs takes more than eyeballing charts. Traders use specific tests to avoid false correlations that look good but fail in live trading.
| Method | What It Tests | Common Threshold | Limitation |
|---|---|---|---|
| Pearson Correlation | Linear price movement similarity | Above 0.80 | Ignores non-linear patterns |
| Cointegration (ADF Test) | Long-term equilibrium between prices | P-value < 0.05 | Slow to detect breaks |
| Distance Method | Normalized price difference volatility | Low historical variance | Misses structural changes |
| Copula Approach | Joint distribution of returns | Time-varying dependence | Computationally complex |
| Minimum Profit Rule | Expected round-trip gain | Exceeds transaction costs | Ignores tail risks |
Many retail traders skip cointegration testing. They lose money on pairs that look correlated but drift apart permanently.
Gold and silver often move together. But in 2020, gold hit records while silver lagged. Traders who assumed automatic reversion lost heavily. The spread can stay irrational longer than you stay solvent.
Once a pair passes tests, traders build a trading model around the spread behavior. The z-score is the workhorse metric here.
| Z-Score Level | Signal | Typical Action | Risk Management |
|---|---|---|---|
| Above +2.0 | Overbought spread | Sell winner, buy loser | Stop at +3.0 |
| Between +1.0 and +2.0 | Mild deviation | Watch, no position | Monitor momentum |
| Between -1.0 and +1.0 | Normal range | No trade | Wait for breakout |
| Between -2.0 and -1.0 | Mild deviation | Watch, no position | Monitor momentum |
| Below -2.0 | Oversold spread | Buy winner, sell loser | Stop at -3.0 |
The z-score measures how far the spread deviates from its 20-day or 60-day average. Higher absolute values mean stronger signals but also higher risk.
A mediocre pair entered at z-score ±2.5 often outperforms a perfect pair entered at ±1.2.
Patience in waiting for extremes separates profitable traders from the rest.
Risk management distinguishes professionals from amateurs. A pair can stay divergent for months. Strategy decay is real and deadly.
| Risk Type | Manifestation | Mitigation Tactic |
|---|---|---|
| Convergence Risk | Spread never reverts | Hard stop at 2x historical maximum |
| Model Breakdown | Cointegration fails suddenly | Real-time ADF monitoring |
| Execution Slippage | Bad fills in volatile legs | Limit orders, small size |
| Regime Change | Industry structure shifts | Quarterly pair review |
| Over-leverage | Too much capital deployed | Capital limit per pair (2% max) |
Capital allocation rules matter enormously. Even a 70% win rate strategy fails with ruinous position sizing.
Long-Term Capital Management used pairs trades with extreme leverage in 1998. Their_models were correct in the long run. Their position sizes killed them in the short run. Survivorship meant staying small enough to wait.
Modern execution uses algorithms to minimize market impact. Execution quality directly affects edge retention after costs.
| Key Point | What It Means | Action Item |
|---|---|---|
| Cointegration over correlation | Shared long-term trend matters more than similar short-term movement | Always run ADF test before trading |
| Z-score extremes signal opportunity | Standard deviation measures how unusual current spread is | Set alerts at ±2.0, act at ±2.5 |
| Position sizing preserves capital | Right model with wrong size still loses everything | Cap single pair exposure at 2% of portfolio |
| Stop losses prevent ruin | Mean reversion is probable, not guaranteed | Hard stop at 3x planned loss or model break |
| Costs erode small edges | Commissions, borrow fees, and slippage compound | Trade liquid pairs, use low-cost brokers |