Definition
Monte Carlo standard error measures the simulation noise in an estimate formed from a finite number of simulated outcomes. For iid outcomes with finite variance, it is estimated by the sample standard deviation of the outcomes divided by the square root of the number of simulations. It is distinct from uncertainty caused by choosing the wrong simulation model.
How it is calculated
Monte Carlo standard error of the simulated mean = sample standard deviation of simulated outcomes divided by square root of M simulations.
| Symbol | Meaning | Unit |
|---|---|---|
| MCSE | Monte Carlo standard error | units of Y |
| Y | simulated outcome | target quantity units |
| s | sample standard deviation | units of Y |
| M | number of simulations | count |
-
Use the variance of the simulated mean under independent draws.
-
Estimate outcome variance with its sample counterpart.
-
Divide the sample standard deviation by the square root of the simulation count.
Worked example
Synthetic example.
-
inputs:
-
sample standard deviation: 2
-
simulation count: 4
-
-
result:
- monte carlo standard error: 1
Interpretation
The simulated average jitters because a finite run sees only some possible outcomes. Four times as many iid simulations halve that typical jitter, which is the square-root law. They do not move an estimate toward truth if the generator is misspecified.
Assumptions and conventions
-
Simulated outcomes are independent or the dependence adjustment is included.
-
M >= 2 when estimating the sample standard deviation.
-
The outcome variance is finite and estimated with a stated sample convention.
-
The standard error concerns simulation error only.
Same units as Y; M is a count.
Limitations and common misreadings
-
MCSE is not model risk or parameter uncertainty.
-
The iid formula needs adjustment for correlated simulation draws.
References
- Paul Glasserman. (2004). Monte Carlo Methods in Financial Engineering. Stochastic Modelling and Applied Probability. Chapter 1, Introduction — Monte Carlo sample-mean estimation, variance and standard error. doi:10.1007/978-0-387-21617-1. Source

