Why Monte Carlo Option Values Use Equal-Weighted Sample Averages
Summary
The document addresses why Monte Carlo option pricing typically averages simulated payoffs equally instead of applying separate probability weights to each simulated outcome. Its central explanation is that simulations drawn from the risk-neutral process already reflect that process’s probability distribution: common outcomes appear more often than tail outcomes, so the sample average approximates the distribution-weighted expected payoff as the number of paths grows.
A second response cautions that adding weights requires choosing a distribution, while Monte Carlo is often used precisely when the process distribution is not available in a directly usable analytical form. The estimate is conditional on the model and parameters used to generate paths; simulation does not guarantee correctness if those assumptions are wrong. The discussion is conceptual and does not cover convergence rates, variance reduction, sampling error, or cases where importance sampling deliberately changes the sampling distribution and then corrects with weights.
Key ideas
- Monte Carlo prices are estimated by averaging payoffs from simulated paths.
- Equal weighting works because the generated paths occur according to the assumed risk-neutral distribution.
- Frequently occurring outcomes contribute more samples than rare tail outcomes.
- The estimate depends on the process model and parameters used to generate the paths.
- The document does not discuss specialized methods that alter sampling probabilities and apply corrective weights.
Tags
Full text
# Why do we use a simple average for pricing options in MonteCarlo? # Why do we use a simple average for pricing options in MonteCarlo? I was recently reviewing my notes on the Binomial Trees and MonteCarlo (MC) methods for option pricing. I've taken this for granted and just used the method....but I started questioning why we take a simple average for calculating the option value. I understand that we're drawing independent random samples from the normal distribution....but isn't the prob of drawing say 0.65 different from drawing 1.465 (etc.)? If so, why don't we do a probability weighted expectation (like in the Binomial Tree) instead of the simple/equally weighted expectation in MC? Would appreciate some guidance. Thanks! ## Answer by KaiSqDist (score 3, accepted) https://quant.stackexchange.com/a/77005 I assume you are talking about taking a probability-weighted average instead of equal-weighted? There is no need for that because naturally a simulated stochastic process will end up more in the regions that is represented by the risk-neutral distribution (mean rather than the tails for a normal distribution for example). Therefore, taking an equal-weighted average "already accounts for the probability-weighting part". ## Answer by THATS MY QUANT MY QUANTITATIVE (score 0) https://quant.stackexchange.com/a/77006 What weighted distribution would you choose? Besides, in situations where an analytical solution is not available, we don't know the distribution of the process, only the distribution of the white noise. If you already know the distribution of the process, you wouldn't bother with a Monte-Carlo simulation - it's computationally expensive. The cool thing about MC is it's "always correct" (in the sense of the parameters you gave it). If you assume some distribution to correct the output, you will be more wrong than the results of the simulation.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.