Monte Carlo simulation is a technique that draws thousands of random scenarios from an assumed model of returns, then looks at the distribution of results. In investing it is used to estimate the range of portfolio outcomes, risk measures and option values.
How it works
- Assume a model: expected return, volatility, and how assets relate.
- Draw random returns for each period and build a path.
- Repeat 10,000 times or more.
- Study the spread of final values, losses or other outputs.
An illustrative example
Rs 1 crore invested for 5 years, with an assumed expected return of 10% and volatility of 15% a year, with lognormally distributed returns:
| Percentile of outcomes | Value after 5 years |
|---|---|
| 5th (poor) | about Rs 90 lakh |
| 50th (median) | about Rs 1.56 crore |
| 95th (good) | about Rs 2.71 crore |
So in this model there is roughly a 5% chance of ending below the starting capital, and a wide range around the median. The numbers are illustrative and come from the assumptions, not from a forecast.
Uses
- Portfolio outcome ranges and goal-based planning.
- Estimating VaR and expected shortfall for non-linear portfolios.
- Pricing options with no closed formula.
- Testing how robust a strategy is by resampling its returns.
Limits
- Garbage in, garbage out. The output reflects the assumed model. Normal or lognormal returns understate crashes.
- Constant volatility and correlations are unrealistic. See GARCH models and market regimes.
- It shows possible outcomes, not probabilities of real events, and it cannot create information the model lacks.
Frequently asked questions
What is Monte Carlo simulation?
It is a method that runs a large number of random scenarios from an assumed model to estimate the distribution of possible outcomes.
How is Monte Carlo used in portfolio analysis?
To estimate the range of future portfolio values, the chance of falling short of a goal, and risk measures such as VaR.
Is a Monte Carlo result a forecast?
No. It shows what could happen under the assumptions given, so the model and its inputs determine the answer.
How many simulations are needed?
Often thousands to hundreds of thousands. More paths reduce sampling noise, especially for tail measures.