On This Page
Key Takeaways
- Monte Carlo estimates a selected quantity from random samples. The output can be a price, probability, quantile or collection of scenario results.
- Choose the modeled variables, their dynamics and dependence, then evaluate the quantity required by the trade or portfolio question.
- For an average of independent, identically distributed finite-variance samples, standard error scales as 1 / √N. Quadrupling N halves that standard error.
- It is one of three standard ways to estimate value at risk, alongside the parametric method and historical simulation.
- Use risk-neutral dynamics for the corresponding option-pricing calculation and the selected return dynamics for a separate investment scenario.
What is Monte Carlo simulation?
Monte Carlo simulation is a method that estimates a quantity by drawing random inputs from a chosen model, evaluating each sample and aggregating the results. In finance it can estimate an option price, a probability, a loss quantile such as value at risk, or a collection of portfolio paths. Its accuracy depends on the model behind the draws and on the number of samples.
The calculation starts with the question. A European terminal payoff needs a terminal-price sample. An average-price, barrier or drawdown calculation also needs the relevant path information.
Formula
Risk-neutral GBM: ST = S0 × exp[(r − q − σ² / 2) × T + σ × √T × Z]. European option estimate = exp(−rT) × average terminal payoff.- S_0 is the current underlying price and S_T is the simulated price at horizon T measured in years.
- r is the continuously compounded risk-free rate, q is continuous dividend yield and σ is annualized log-return volatility. The example sets q = 0.
- Z is a standard normal random draw, one per simulated path.
- The exponential form comes from assuming the underlying follows geometric Brownian motion, which produces a lognormal price distribution.
- For independent sampled discounted payoffs with finite variance, estimate the standard error of their average as sample payoff standard deviation divided by √N.
Monte Carlo option pricing example: a European call
Value a one-year European call on a non-dividend-paying stock with spot $100, strike $105, annualized volatility 25% and continuously compounded risk-free rate 2%. Compare a 100,000-draw risk-neutral simulation with its Black-Scholes benchmark.
- Drift term: (0.02 − 0.5 × 0.25²) × 1 = −0.01125
- Diffusion term: 0.25 × √(1) × Z, so a path with Z = 1.0 gives S_T = 100 × exp(−0.01125 + 0.25) = $126.97
- At Z = 1, the payoff is $126.97 − $105 = $21.97 per share, or $2,197 for a standard 100-share contract, before discounting.
- The closed-form benchmark is $8.70 per share. The independently reproduced 100,000-draw run estimates $8.78, with sampling standard error $0.05 per share.
- Raising the run count from 10,000 to 160,000 paths cuts the standard error to roughly one quarter of its prior level
The reproduced run estimates $878 for a standard contract; the closed-form benchmark is $870. Both use the same risk-neutral pricing inputs. Reprice at higher and lower volatility to see how the trade’s value responds.
At Z = 1, terminal price is $126.97 and call payoff is $21.97 per share. Discount the payoff by exp(−0.02) and average the discounted results across draws.
The closed-form Black-Scholes benchmark is $8.70 per share. The reproduced 100,000-draw example gives $8.78 with sampling standard error $0.05. For a standard 100-share contract, the corresponding values are $870 and $878.
How to read the model’s price distribution
The chart’s bars are analytic probabilities for the displayed $10 terminal-price intervals, rounded to one decimal percentage point. The risk-neutral probability above the $105 strike is 40.5%.
The $100 reference is the current price; model terminal mean is $102.02. These are pricing probabilities; a return scenario runs on real-world dynamics.
Monte Carlo standard error and the number of paths
For an average of independent, identically distributed discounted payoffs with finite variance, estimated standard error is sample standard deviation divided by √N. Moving from 10,000 to 160,000 draws divides this standard error by four.
A sampling confidence interval concerns the estimated model quantity. An outcome interval concerns the distribution of prices or losses. Increasing draw count improves the estimate of that distribution without shrinking its underlying spread.
Antithetic samples, control variates and appropriately weighted importance samples provide other estimators. Use the variance calculation belonging to the estimator actually used.
Monte Carlo vs historical simulation and path choices
Historical scenarios apply selected factor changes to current positions. Bootstrapping draws from an empirical distribution and can construct new sequences from those observations.
American-style exercise can be evaluated with lattices or least-squares Monte Carlo. Multi-asset simulation uses the chosen dependence structure across the variables.
A continuous process has no instantaneous jumps, but observations separated by an overnight interval can still differ substantially. Distinguish a discontinuity in the process from a gap between recorded prices.
For a trading position, revalue each holding across the chosen prices, volatility inputs and dates. The Options Profit Calculator maps an options position’s profit and loss across price, date and implied volatility, and portfolio simulation then aggregates the positions for the selected risk or valuation question.
Limitations of Monte Carlo simulation
- The result is only as good as the model. Wrong dynamics, volatility or dependence produce precise-looking wrong numbers, and more draws cut only the sampling error, so rerun the model at other volatility and dependence inputs before you trade on the price.
- Budget draws for the precision you need. For independent finite-variance draws, standard error falls with 1 / √N, so halving it takes four times as many draws.
- Tail quantiles and rare events need many draws or variance reduction, because few samples land in the region that matters.
- Risk-neutral probabilities are pricing numbers: the 40.5% above the $105 strike prices the call. For a return scenario, rerun the simulation with real-world dynamics.
- Path-dependent payoffs need time steps, and coarse steps add discretization error, so check that the price holds when you halve the step.
Bottom line
Monte Carlo simulation turns a model into numbers by sampling it many times, which makes it a flexible tool for path-dependent payoffs, portfolio risk and questions with no closed-form answer. Set a precision target, report the standard error, and remember that more draws sharpen the estimate but never fix the model.
Clear answers
Frequently Asked Questions
Sources and Further Reading
- CFA Institute: Backtesting and Simulation
- CFA Institute: Measuring and Managing Market Risk
- Federal Reserve (FEDS): How Accurate Are Value-at-Risk Models at Commercial Banks?
- OIC: The Black-Scholes Formula
How entries are sourced, calculated and updated: editorial methodology.