IV ExpertWeek 21 • Lesson 62Duration: 35 min

SRM Statistical Arbitrage Risk Management

The five risks that will kill your stat arb P&L — and how to monitor each one

Learning Objectives

•Identify the specific risks unique to statistical arbitrage
•Build a comprehensive risk monitoring framework
•Implement automated risk checks for production stat arb

Explain Like I'm 5

Stat arb has a specific set of risks that directional trading doesn't. Crowding risk (everyone running the same trade). Correlation breakdown (the relationship stops working). Gap risk (overnight jumps that bypass stops). Margin risk (two positions burning margin). Model risk (your OU or Kalman filter is wrong). Each needs its own monitoring and defense. This lesson builds the complete risk framework.

Think of It This Way

Managing stat arb risk is like running a hospital ICU. Each patient (trade) has multiple vital sign monitors — heart rate (correlation), blood pressure (spread deviation), breathing (liquidity), and brain activity (model confidence). Any single vital sign going critical triggers an alarm. You don't wait for all five to go red simultaneously.

1The Five Stat Arb Risks

These are the specific risks that will catch you: 1. Crowding risk. When too many participants run the same stat arb strategy, returns compress and drawdowns become correlated. The August 2007 quant crisis was exactly this — the same stat arb factors unwindled simultaneously. 2. Correlation/cointegration breakdown. The statistical relationship you're trading can change permanently. Central bank policy shifts, regulatory changes, or structural market changes can break relationships that worked for years. 3. Gap risk. Weekend gaps or flash crashes can create massive spread divergence that bypasses any intraday stop-loss. Since you have TWO positions, both can gap against you. 4. Margin risk. Two positions = double the margin usage. In drawdown, margin requirements can increase while your equity is shrinking. This creates a forced-liquidation death spiral. 5. Model risk. Your OU model, Kalman filter, or cointegration test could be systematically wrong. Over-fitted parameters, stale calibration, or structural breaks in the data-generating process all create model risk.

2Risk Monitoring Framework

For each of the five risks, you need a quantitative monitor: Crowding monitor. Track: - Short interest on your instruments - Volume spikes that suggest informed trading - Correlation of your returns with known factor indices Relationship monitor. Track: - Rolling cointegration p-value (trigger at p > 0.10) - Rolling correlation (trigger at 20% decline from baseline) - Spread half-life drift (trigger if half-life doubles) Gap risk monitor. Track: - Weekend gap history on both legs - Upcoming economic events (NFP, FOMC, ECB) - Reduce exposure before known risk events Margin monitor. Track: - Current margin utilization (trigger at 60%) - Projected margin if spread widens 50% - Keep reserve margin for adverse scenarios Model monitor. Track: - Prediction residuals over time (should be white noise) - Parameter stability (kappa, mu, sigma drift) - Out-of-sample performance vs. in-sample Build a dashboard that shows all five monitors in real-time. Any single red flag should trigger position reduction. Two or more red flags should trigger a full unwind.

Key Formulas

Max Divergence Stop-Loss

A hard stop-loss on the spread at 4 standard deviations. Beyond this, the probability that the spread reverts within a reasonable timeframe is too low. Cut the loss.

Hands-On Code

Stat Arb Risk Monitor

python

import numpy as np

def statarb_risk_check(spread, coint_pvalue, margin_used_pct, half_life_current, half_life_baseline):
    """Comprehensive stat arb risk assessment."""
    alerts = []
    
    # Spread deviation check
    z = (spread[-1] - np.mean(spread[-100:])) / np.std(spread[-100:])
    if abs(z) > 4.0:
        alerts.append(f"[CRITICAL] Spread z={z:.1f} exceeds 4-sigma stop")
    elif abs(z) > 3.0:
        alerts.append(f"[WARNING]  Spread z={z:.1f} approaching stop zone")
    
    # Cointegration stability
    if coint_pvalue > 0.15:
        alerts.append(f"[CRITICAL] Cointegration p={coint_pvalue:.3f} (lost)")
    elif coint_pvalue > 0.08:
        alerts.append(f"[WARNING]  Cointegration p={coint_pvalue:.3f} (weakening)")
    
    # Margin check
    if margin_used_pct > 0.70:
        alerts.append(f"[CRITICAL] Margin utilization {margin_used_pct:.0%}")
    elif margin_used_pct > 0.50:
        alerts.append(f"[WARNING]  Margin utilization {margin_used_pct:.0%}")
    
    # Model stability (half-life drift)
    hl_ratio = half_life_current / max(half_life_baseline, 1)
    if hl_ratio > 2.5:
        alerts.append(f"[CRITICAL] Half-life {hl_ratio:.1f}x baseline (model stale)")
    elif hl_ratio > 1.5:
        alerts.append(f"[WARNING]  Half-life {hl_ratio:.1f}x baseline")
    
    # Summary
    criticals = sum(1 for a in alerts if 'CRITICAL' in a)
    warnings = sum(1 for a in alerts if 'WARNING' in a)
    
    print(f"=== STAT ARB RISK CHECK ===")
    for alert in alerts:
        print(f"  {alert}")
    if not alerts:
        print(f"  [ALL CLEAR] No risk flags")
    print(f"")
    print(f"Criticals: {criticals}  Warnings: {warnings}")
    
    if criticals >= 2:
        print(f"  >>> UNWIND ALL POSITIONS <<<")
    elif criticals >= 1:
        print(f"  >>> REDUCE EXPOSURE 50% <<<")
    elif warnings >= 2:
        print(f"  >>> TIGHTEN STOPS <<<")
    
    return criticals, warnings

Implements a comprehensive risk check that evaluates cointegration stability, spread deviation, margin utilization, and model health to produce an aggregate risk assessment.

Knowledge Check

Q1.Your pair's cointegration p-value rises from 0.01 to 0.15 over two weeks. What should you do?

Assignment

Build a comprehensive risk monitoring dashboard for a stat arb portfolio. Implement automated alerts for all five risk types. Backtest your risk system against the 2020 COVID crisis to validate it would have protected capital.

← Previous Lesson Next Lesson →