Futures Statistical Arb

Regular trading bets on market direction (up or down). Stat arb bets on the relationship between two instruments - it's market neutral. You profit if the spread normalizes, regardless of whether the overall market goes up or down.

Best pairs have high correlation (0.70+), are cointegrated (ADF test passes), have reasonable half-life (2-15 days), and have a fundamental economic link. Examples: ES/NQ (both US indices), GC/SI (both precious metals), ZB/ZN (both Treasuries).

Stat arb requires enough capital to trade multiple contracts of two instruments simultaneously. With futures margin, $50,000 minimum is recommended for one pair. For proper diversification across 3-5 pairs, $150,000+ is better.

Main risks: Spread divergence (spread widens instead of reverting), regime change (correlation breaks), leg risk (one leg fills poorly), and model risk (statistics fail). Manage with Z-score stops, time stops, correlation monitoring, and diversification.

Trade duration depends on half-life. Typical stat arb trades last 3-15 days. Fast half-life pairs (3-5 days) can be day traded. Slower pairs (10-20 days) are swing trades. Very long half-life (30+ days) is usually impractical.

Common methods: OLS regression (regress Leg1 on Leg2, beta is hedge ratio), rolling regression (30-60 day window), or Kalman filter (dynamic). For dollar neutrality, also adjust for contract multipliers. Test which method gives most stationary spread.

If ADF test fails (p-value > 0.05), the relationship may have broken. Exit existing positions, remove pair from trading, and investigate cause. Retest after 30-60 days to see if relationship restores. Some breakdowns are temporary.

Contract rollover affects spreads differently. Some pairs (like index spreads) roll similarly. Others (like calendar spreads) are specifically trading the roll. Always check spread behavior around roll dates and adjust positions or pause if necessary.

Z-score stops (e.g., exit at ±3.0) are more appropriate for stat arb because they're relative to the spread's statistical distribution. Price stops on individual legs ignore the spread relationship and can trigger on normal leg movement.

Trade multiple pairs with low correlation to each other. An ES/NQ position shouldn't correlate with a GC/SI position. Allocate based on volatility (risk parity) or equal weight. Target 5-10 pairs for proper diversification.

Use Kalman filter when the hedge ratio is clearly time-varying (relationship evolving) or when you need smooth adaptation. Use OLS when relationship is stable or for simplicity. Kalman is better for live trading; OLS is fine for backtesting.

Use CUSUM test or Chow test for structural break detection. Also monitor rolling cointegration p-values - if they trend toward 0.05, relationship is weakening. Compare 30-day vs 90-day correlations. Sudden divergence suggests break.

Use mean-variance optimization with shrinkage estimators, or risk parity for more stable weights. Consider pair correlations (avoid concentrated exposure). Rebalance monthly or when weights drift significantly. Account for transaction costs.

Model failure causes: Regime change (crisis), fundamental change (one company acquired), changing correlation structure, overfitting historical data, or extreme leverage. Mitigate with robust testing, out-of-sample validation, and conservative sizing.

Decompose returns into: pair selection (which pairs traded), timing (entry/exit Z-scores), execution (fill quality vs theoretical), and costs. Track Sharpe per pair and overall. This identifies what's working and what needs improvement.