Monte Carlo Options Pricing

The Monte Carlo option pricing method values options by simulating thousands (or millions) of possible price paths from today to expiration date, calculating the option payoff on each path, and averaging to find expected value. Monte Carlo is particularly suited to exotic options (asian-option, barrier-option) with path-dependent payoffs that Black-Scholes model cannot handle analytically. It is more flexible than binomial-option-pricing but computationally intensive.

How Monte Carlo works

1. Generate random price paths: Using a stochastic process (usually geometric Brownian motion), simulate N price paths from today to expiration.

2. Calculate payoff on each path: For each final price (or path-dependent value), compute the option payoff.

3. Average payoffs: Take the mean of all N payoffs.

4. Discount to present: Discount the expected payoff by the risk-free rate.

Formula:

Option Value = e^(−r×T) × (1/N) × Σ payoff_i

Example: Asian option

An asian-option call struck at $100 requires averaging the price over the option’s life.

• Simulate 10,000 price paths
• On each path, calculate the average price
• Payoff = max(average − 100, 0)
• Average all 10,000 payoffs
• Discount by risk-free rate

This can be done analytically for some options (Black-Scholes model) but Monte Carlo handles any averaging rule.

Stochastic processes

The most common process is geometric Brownian motion:

dS = μ × S × dt + σ × S × dW

Where:

• S is the stock price
• μ is the drift (expected return)
• σ is volatility
• W is a Wiener process (random)

This generates realistic price paths with log-normal returns.

Discretization

To simulate, the path is discretized into small time steps:

S_new = S_old × e^((r − 0.5σ²)Δt + σ√Δt × Z)

Where Z is a standard normal random variable.

Convergence and accuracy

Monte Carlo has slow convergence: error decreases as O(1/√N). To halve the error, you need 4x more simulations.

With 10,000 simulations, error is ~√(1/10,000) = 1%. With 1,000,000, error is ~0.1%.

Variance reduction techniques (antithetic variates, control variates) speed convergence without more simulations.

Quasi-random numbers

Standard random numbers have clustering inefficiencies. Quasi-random (low-discrepancy) sequences (Sobol, Halton) fill the space more evenly, improving convergence.

Quasi-random often gives 5–10x faster convergence than random.

Greeks via Monte Carlo

Computing Greeks (delta, gamma, vega) via Monte Carlo requires either:

• Bumping: Run simulations with slightly perturbed inputs; differences approximate Greeks.
• Pathwise derivatives: Some payoffs have analytical derivatives usable directly.

Bumping is simple but expensive (requires re-running all simulations).

Applications

Exotic options: Basket, asian-option, barrier, lookback, rainbow options.

Stochastic volatility: Models where volatility itself changes randomly.

Jump diffusion: Stock prices can jump (gaps overnight).

Multi-asset: Options on multiple underlying assets with correlation.

Advantages and disadvantages

Advantages:

• Flexible; handles any payoff
• Intuitive; simulates reality
• Parallelizable; runs on GPUs

Disadvantages:

• Slow convergence (need many simulations)
• Early exercise (American options) complex to handle
• More computationally expensive than Black-Scholes

See also