Monte Carlo Simulation

Monte Carlo simulation estimates financial results by repeatedly drawing uncertain inputs from a specified probability model and calculating an outcome for each draw. It can estimate an average value, a loss probability, or a distribution of possible cash flows.

The method calculates what follows from the model. It does not establish that the model represents the future correctly.

Key Takeaways

• Inputs need distributions, dependence assumptions, and a defined time horizon.
• Each simulated outcome must use the same financial rules, units, and cash-flow timing.
• More runs generally reduce sampling error, not errors in assumptions or implementation.
• Derivative pricing requires the appropriate pricing probabilities and discounting, not simply an average of forecast payoffs.

How a Simulation Works

1. Define the output. For example, estimate one-year project NPV or the probability that a portfolio loss exceeds a stated threshold.
2. Specify uncertain inputs. Document distributions, parameters, and relationships between variables. Drawing each input independently is itself an assumption.
3. Generate draws or paths. A path is needed when timing or intermediate values affect the result.
4. Calculate the outcome for each draw. Apply costs, contractual payoffs, cash-flow timing, and discounting consistently.
5. Summarize and check. Report relevant means, percentiles, loss frequencies, and numerical uncertainty. Compare with a simpler calculation where possible.

For equally weighted draws, an estimated mean is:

Here, N is the number of draws and Y is the financial output for each draw, such as a project’s NPV. An average does not describe every adverse outcome. MIT’s Monte Carlo lecture notes introduce the underlying sampling method.

Worked Example: Simulating a One-Year Project

Assume a hypothetical project costs USD 100,000 today and produces one net cash inflow at year-end. The simplified model assigns a 20% probability to each of five possible inflows. Use a 10% annual discount rate solely for this illustration, with no other cash flows.

Suppose ten demonstration draws produce these counts:

The sampled inflows total USD 1,110,000, so their mean is USD 111,000. Because the discount rate and initial cost are fixed, the estimated mean NPV is:

An inflow below USD 110,000 produces a negative NPV. Five of the ten draws meet that condition, so the sample’s estimated probability of negative NPV is 50%.

This tiny model can also be calculated exactly. Its probability-weighted mean inflow is USD 112,000, its expected NPV is approximately USD 1,818.18, and its probability of negative NPV is 40%. The ten-draw results differ because of sampling variation.

Ten draws are not presented as a reliable estimation standard. The example makes the arithmetic visible; simulation becomes more useful when many interacting inputs or paths make direct enumeration impractical.

Sampling Error Is Not Model Error

For independent, identically distributed outputs with finite variance, the estimated standard error of the sample mean is:

Here, s_Y is the sample standard deviation of the simulated outputs. Under these conditions, approximately four times as many draws are needed to halve the standard error. See MIT’s sampling and standard-error notes for the statistical basis.

This is uncertainty in estimating the model’s mean, not the spread of future financial outcomes. It does not cover incorrect distributions, uncertain parameters, coding errors, or omitted risks. A larger run also does not guarantee that every individual estimate moves closer to the correct model value.

Rare-event probabilities and tail percentiles need their own error assessment. Zero observed losses in a finite sample do not prove that loss is impossible. Techniques such as importance sampling may help, but changing the sampling distribution requires the correct weights. Columbia’s Monte Carlo methods notes explain this distinction and rare-event estimation.

Derivative Pricing: Discount the Right Expected Payoff

For a European call in a simplified model with a constant continuously compounded risk-free rate, a Monte Carlo pricing estimate takes the form:

The terminal stock prices S must be drawn under the model’s risk-neutral measure; K is the strike, T is time to expiration in years, and r uses the same time basis. The exponential factor discounts the payoff to today.

Averaging payoffs generated with an analyst’s forecast stock return is not automatically an arbitrage-consistent option price. Dividends, stochastic rates, early exercise, and other contract features require appropriate treatment. See Broadie and Detemple’s option-pricing survey and the site’s risk-neutral guide below.

Common Mistakes

• Treating simulated percentiles as guaranteed limits on loss.
• Ignoring dependence, changing market conditions, or outcomes excluded by the distribution.
• Reporting many decimal places while input estimates remain highly uncertain.
• Confusing more simulation paths with a finer time grid; time-discretization error is separate.
• Omitting fees, withdrawals, taxes, defaults, or contract features when they matter to the stated use.
• Failing to record the model version, assumptions, and random seed needed to reproduce a run.

Related Terms

• Stochastic Modeling: The probability model that a simulation evaluates.
• Risk-Neutral Probabilities: Pricing weights distinct from real-world forecasts.
• Scenario Analysis: Coherent alternative conditions, with or without probability weights.
• Sensitivity Analysis: How input choices affect a result.
• Model Risk: Incorrect or inappropriate reliance on a model.

Check Your Understanding

FAQs

This article provides financial-modeling education, not investment advice or a guarantee of returns, liquidity, or forecast accuracy.