Statistical Arbitrage Strategy: Mathematical Formulation & Implementation

Academic & Industry Origins

Statistical Arbitrage evolved in the mid-1980s at Morgan Stanley under the leadership of Gerry Bamberger and David Shaw, pioneering quantitative pairs trading. In 1987, Robert Engle and Clive Granger introduced the mathematical concept of cointegration—a discovery that earned the 2003 Nobel Prize in Economics—providing the theoretical foundation for testing whether a linear combination of non-stationary time series is stationary.

Economic Mechanism

Companies sharing identical sector exposures, revenue models, and regulatory environments (e.g. major payment processors or commercial banks) are subjected to the same macro cash flow factors. When short-term supply-demand shocks drive their relative prices apart, market-neutral capital enters to capture the expected convergence back to historical fair value.

Strengths & Limitations

• Strengths: Generates returns with low correlation to broader equity market indices; operates effectively during range-bound and sideways market regimes.
• Limitations: Vulnerable to permanent structural breaks (mergers, regulatory actions, catastrophic single-company fraud); requires low-cost borrow availability and tight execution synchronization.

The Statistical Arbitrage engine rigorously tests and trades cointegrated spreads:

1. Cointegration Regression (Engle-Granger Two-Step)

For candidate equity pair $(P_A, P_B)$, an ordinary least squares (OLS) regression estimates the hedge ratio $\beta$:

$$P_{A,t} = \alpha + \beta P_{B,t} + \epsilon_t$$

Where $\epsilon_t$ represents the residual spread series:

$$\epsilon_t = P_{A,t} - (\alpha + \beta P_{B,t})$$

2. Augmented Dickey-Fuller (ADF) Stationarity Test

The residuals $\epsilon_t$ must reject the null hypothesis of a unit root at a 95% or 99% confidence level ($p < 0.05$):

$$\Delta \epsilon_t = \gamma \epsilon_{t-1} + \sum_{i=1}^p \delta_i \Delta \epsilon_{t-i} + u_t$$

If $\gamma < 0$ and statistically significant, the spread is verified as mean-reverting.

3. Standardized Spread Z-Score

The residual spread is normalized over lookback window $N$ (default $N = 60$ trading days):

$$\mu_\epsilon = \frac{1}{N} \sum_{i=0}^{N-1} \epsilon_{t-i}, \quad \sigma_\epsilon = \sqrt{\frac{1}{N} \sum_{i=0}^{N-1} (\epsilon_{t-i} - \mu_\epsilon)^2}$$

$$Z_{\text{spread}, t} = \frac{\epsilon_t - \mu_\epsilon}{\sigma_\epsilon}$$

4. Estimated Half-Life of Spread Reversion

The half-life of reversion governs expected holding duration:

$$t_{1/2} = -\frac{\ln(2)}{\gamma}$$

Signals are triggered when the spread dislocates beyond predefined standard deviations:

graph TD
    A[Monitor Stationary Residual Spread Z-Score] --> B{Spread Z-Score Evaluation}
    B -->|Z_spread < -2.0| C[LONG Spread: Buy Asset A, Short Asset B]
    B -->|Z_spread > +2.0| D[SHORT Spread: Short Asset A, Buy Asset B]
    B -->|abs Z_spread <= 0.5| E[NEUTRAL: Close Position at Convergence]
    B -->|abs Z_spread >= 3.5| F[CUTOFF: Immediate Emergency Liquidation]

Operational Rules

1. Long Spread Entry (Z_spread < -2.0): Asset A is significantly undervalued relative to Asset B. The system issues a LONG directive for Asset A and a paired SHORT hedge for Asset B.
2. Short Spread Entry (Z_spread > +2.0): Asset A is significantly overvalued relative to Asset B. The system issues a SHORT directive for Asset A and a paired LONG hedge for Asset B.
3. Convergence Exit (|Z_spread| <= 0.5): When the spread reverts to within 0.5 standard deviations of historical mean, the paired position is fully liquidated at profit.
4. Structural Break Cutoff (|Z_spread| >= 3.5): If the spread widens beyond 3.5 standard deviations, the statistical relationship is deemed broken (e.g. unexpected litigation or earnings breakdown). The trade is immediately closed to eliminate unbounded losses.

• Leg Synchronization: Execution must execute both legs simultaneously using limit or pegged orders to prevent execution slippage from destroying the spread margin.
• Corporate Action Tracking: Cash dividends, stock splits, and spinoffs directly alter the nominal spread value and must be dynamically adjusted in the historical regression series.
• Short Borrow Availability: Asset B must be verified as Easy-To-Borrow (ETB) with minimal borrow interest rates to avoid negative financing carry.