📊 Cointegration & Stat Arb Scanner

What is Cointegration?

Cointegration identifies pairs that share a long-run equilibrium. Unlike correlation (which measures co-movement), cointegration means the spread between two assets is mean-reverting — when it deviates, it tends to return to the mean.

Why it matters: Two assets can be highly correlated but NOT cointegrated (they drift apart over time). Cointegrated pairs are statistically proven to revert, making them ideal for pairs trading.

Engle-Granger Method

We use the Engle-Granger two-step approach:

1. Step 1: Estimate hedge ratio (β) via OLS regression: A = α + β × B + ε
2. Step 2: Test residuals (spread) for stationarity using ADF test

If the spread is stationary (ADF p-value < 0.05), the pair is cointegrated.

Key Metrics

• p-value: Probability spread is NOT stationary. Lower = stronger cointegration.
• Hedge Ratio (β): For every 1 unit of A, trade β units of B in opposite direction.
• Half-Life: Expected time for spread to revert 50% toward mean. Shorter = faster mean reversion.
• Z-Score: Current spread in standard deviations from mean. |Z| > 2 = potential signal.

Trading Signals

• Z-Score < -2: Spread is oversold → Long A, Short B (expect spread to widen)
• Z-Score > +2: Spread is overbought → Short A, Long B (expect spread to narrow)
• Exit: When Z-Score returns to 0 (mean reversion complete)
• Stop-Loss: If |Z-Score| > 3-4 and continuing to diverge

Best Practices

• Only trade cointegrated pairs (p-value < 0.05)
• Prefer half-life between 12-72 hours (fast enough to trade, not noise)
• Use rolling cointegration tests — relationships can break down
• Consider transaction costs — spread must be wide enough to profit after fees
• Size positions to match hedge ratio exactly for market-neutral exposure