While Black-Scholes gives closed-form prices for vanilla options, exotic payoffs — Asian, barrier, lookback — often have no analytical solution. Monte Carlo simulation provides a flexible, model-agnostic framework for pricing any path-dependent derivative.
The Algorithm
For a European-style payoff under risk-neutral GBM:
- Simulate N paths of the underlying from t = 0 to T, using:
S(t+Δt) = S(t) · exp[(r − σ²/2)Δt + σ√Δt · Z]
where Z ~ N(0,1) is a standard normal random draw. - Compute the payoff for each path, e.g. for a call: max(S_T − K, 0).
- Average the payoffs across all N simulations.
- Discount by e^(−rT) to get the present value.
Path-Dependent Payoffs
The real power of MC is pricing path-dependent exotics:
Asian Options
Payoff depends on the average spot price over the life of the option: max(Avg(S) − K, 0). The average can be arithmetic, geometric, or harmonic, and the strike can be fixed or floating. MC naturally handles all variants by accumulating the running average along each path.
Barrier Options
Activated or extinguished when spot crosses a barrier level H. A knock-out call, for example, pays nothing if spot ever breaches H during the option's life. MC checks each time step for barrier crossing.
Caveat: discrete monitoring (checking daily vs. continuously) introduces a bias. The Brownian bridge correction or more frequent time steps mitigates this.
Lookback Options
Payoff depends on the maximum or minimum spot over the life: S_T − min(S_t) for a floating-strike lookback call. The global minimum is tracked along each simulated path.
Convergence & Error
The MC standard error decays as σ / √N — halving the error requires four times as many simulations. For a typical equity option with σ ≈ 20%, achieving accuracy within $0.01 on a $100 strike option requires roughly:
N ≈ (σ_payoff / ε)² ≈ (5 / 0.01)² = 250,000 paths
Variance Reduction Techniques
Antithetic Variates
For every random path using Z, also simulate the path using −Z. The two paths are negatively correlated, reducing variance without extra random draws. Roughly halves the standard error at no extra computational cost.
Control Variates
If a closely related option has a known analytical price (e.g., a geometric Asian has a closed form), simulate it alongside the target. Use the analytical price to correct the MC estimate. This can reduce standard error by 90% or more.
Quasi-Monte Carlo (QMC)
Replace pseudo-random numbers with low-discrepancy sequences (Sobol, Halton). QMC achieves convergence of O(log(N)^d / N) vs. O(1/√N) for standard MC in low dimensions, dramatically improving accuracy for simple payoffs.
Practical Considerations
- Time steps. More steps improve path accuracy but increase computation. For barrier options, use at least daily steps.
- Parallelisation. MC is embarrassingly parallel — each path is independent. GPU acceleration (CUDA) can price complex exotics in milliseconds.
- Greeks via MC. Delta and other Greeks can be estimated via finite difference (bump-and-reprice) or likelihood ratio / pathwise methods which are unbiased but noisier.