Skip to content

Monte Carlo Methods

Difficulty beginner

Overview

Monte Carlo methods use random sampling to solve deterministic problems. In trading, they're essential for pricing complex derivatives, risk analysis, strategy testing, and portfolio optimization.

Core Principle

Estimate quantity by averaging many random samples:
θ ≈ (1/N) Σ f(Xᵢ)  where Xᵢ ~ distribution

Convergence rate: O(1/√N)

where: θ quantity being estimated (an expectation) · f(X) function evaluated at random draw X · N sample size · convergence rate is dimension-free — unlike grid-based integration which is O(N^{-1/d}). does: the law-of-large-numbers estimator — averaging i.i.d. evaluations approximates the underlying expectation. The dimension-free O(1/√N) rate is why MC dominates for high-dimensional pricing and risk problems.

Variance Reduction Techniques

1. Antithetic Variates

For each random draw Z, also use its complement −Z. The two correlated estimates have the same mean but opposite errors, so averaging them halves variance for any monotonic payoff. Free to implement (no new draws to generate) — almost always worth doing for Black-Scholes-style path simulations.

Reduces variance when function is monotonic.

2. Control Variates

Use a correlated variable with known expectation. If you want E[X] and you have a Y strongly correlated with X whose E[Y] is known analytically, the estimator X̂ + β(E[Y] − Ŷ) has smaller variance than X̂ alone. Standard trick in exotic-option pricing: use the analytical European price as a control while simulating the path-dependent payoff.

3. Importance Sampling

Sample from a different distribution that emphasizes important regions, then reweight. Critical for rare-event estimation — VaR at 99.9%, default probabilities, out-of-the-money option pricing — where vanilla MC almost never lands in the region you care about. Tilt the proposal toward the tail and divide by the likelihood ratio to keep the estimator unbiased.

4. Stratified Sampling

Divide sample space into strata and sample from each in proportion to its mass. Guarantees coverage of every region instead of letting random chance under-represent corners of the distribution. Combined with quasi-MC, the workhorse approach for high-dimensional path-dependent pricing.

Applications in Trading

Sample paths visualized

price S₀ T (time) mean E[Sₜ] upside tail downside tail
20 sample paths of geometric Brownian motion · thick line = mean trajectory · fan width grows with √t (variance scales linearly with time)

Quasi-Monte Carlo

Uses low-discrepancy sequences instead of random samples:

Faster convergence: O((log N)^d / N) vs. O(1/√N)

Practical Guidelines

Problem Recommended Method
Simple integration Standard MC
Monotonic function Antithetic variates
Known benchmark Control variates
Tail estimation Importance sampling
High-dimensional Quasi-MC
Path-dependent options Standard MC with many paths

Key Considerations

  1. Number of Paths — More paths = better accuracy, but diminishing returns
  2. Random Seed — Set for reproducibility
  3. Convergence Check — Monitor estimate stability
  4. Variance Reduction — Always consider before increasing paths
  5. Parallelization — Monte Carlo parallelizes trivially

Key Formulas Reference

MC Estimate: θ̂ = (1/N) Σ f(Xᵢ)
Standard Error: SE = σ_f / √N
95% CI: θ̂ ± 1.96 × SE
Convergence Rate: O(1/√N)
GBM Path: S_{t+1} = S_t exp((μ-σ²/2)Δt + σ√Δt Z)

Next Steps