What are the key simulation techniques and variance reduction methods used in risk management?

Monte Carlo simulation estimates risk metrics by averaging across thousands of random scenarios. The problem: with limited simulations, estimates are noisy. Variance reduction techniques improve precision without running more simulations.

Basic Monte Carlo for VaR:

1. Model the return process (e.g., GBM for each asset)
2. Generate N random scenarios (say N = 10,000)
3. Calculate portfolio value in each scenario
4. Sort losses from largest to smallest
5. VaR at 99% = the 100th largest loss

The Noise Problem:

Standard error of a Monte Carlo estimate decreases as 1/sqrt(N). To halve the error, you need 4x the simulations. For tail quantiles (VaR, ES), the noise is even worse because few observations fall in the tail.

Variance Reduction Techniques:

1. Antithetic Variates:

For every random draw Z, also use -Z. Since Z and -Z are perfectly negatively correlated, their average has lower variance.

Procedure:

• Draw Z1 = 0.8 -> compute return R1
• Use -Z1 = -0.8 -> compute return R1*
• Average estimate = (R1 + R1*)/2

This works especially well when the payoff is monotonic in Z (most financial applications).

2. Control Variates:

Use a variable whose expected value you know analytically to reduce noise.

Example: Pricing an exotic option. You also simulate a vanilla option (whose price you know from Black-Scholes). The error in the vanilla estimate gives you information about the error in the exotic estimate.

Adjusted estimate = MC_exotic - beta x (MC_vanilla - BS_vanilla)

3. Importance Sampling:

Shift the probability distribution to generate more tail events. Then reweight each scenario to correct for the shift.

For VaR estimation, shift the mean of the return distribution toward negative values -> more extreme losses are generated -> tail estimate is more precise.

Practical Comparison at Westbrook Quant Desk:

FRM Exam Focus:

• Know the intuition behind each method
• Antithetic variates: simple, always helps for monotonic payoffs
• Control variates: powerful when a good control exists
• Importance sampling: best for tail estimation but can backfire if poorly calibrated

Master simulation techniques in our FRM Part I course.