Statistical Arbitrage: Signal Extraction & Risk Overlays

Statistical arbitrage is the generalized version of pairs trading. Instead of trading a single cointegrated pair, stat-arb trades hundreds or thousands of weak mean-reversion signals across a large equity universe, relying on diversification rather than per-trade conviction.

Core mechanic

1. Define a residual model. Strip systematic exposure (market beta, sector, factor) from individual stock returns. The residual is what you trade.
2. Score residuals for mean-reversion. Common signals: short-horizon residual returns (1–5 day), order flow imbalance, post-earnings drift, news residuals.
3. Construct a beta- and factor-neutral portfolio weighted by signal strength.
4. Rebalance frequently — daily to intraday — to harvest the short-horizon edge.

Why the risk overlay matters more than the signal

The signal in stat-arb is statistically real but economically weak. A stat-arb portfolio's per-name edge is fractions of a basis point per trade. What makes the portfolio investable:

• Factor neutrality. Long-short market exposure is hedged. Sector exposure is hedged. Style factors (value, growth, size) are hedged.
• Risk-controlled aggregation. Position limits, gross-exposure caps, per-name caps.
• Liquidity discipline. Position size as % of average daily volume, not capital fraction.

Strip the overlays and a stat-arb book is a beta-1 long-short equity fund with a high turnover. The overlays are the strategy.

Failure modes

• 2007 quant crisis. Multiple stat-arb books deleveraged simultaneously, forcing positions in the same direction. Diversification across signals collapsed.
• Regime change in microstructure. When tick size, spread structure, or HFT dominance changes, signals calibrated to the old regime stop working.
• Capacity decay. Stat-arb capacity is finite. Every dollar deployed compresses the edge for the dollars that follow.

Related

• Pairs trading
• Mean reversion
• Correlation decay