- Basic
Let
Y
denote a continuous random variable. Using Monte Carlo method, we first consider the integral
with variance
For large sample size, n−1 in equation above is replaced by n. By the Central Limit Theorem, we have
Therefore, we know that in order to reduce the standard error of ĝ ,which equals ŝ 2n‾‾‾√ , we need either increase the number of trials n or reduce the size of
- Variance Reduction Techniques: Control Variate method
For a given n,
generate yi from g(y)
generate Yi from h(y)
ĝ = ∫g(y)f(y)dy
Ĝ =∫g(y)h(y)dy
g∗=G+(ĝ −Ĝ ) , which is a revised estimate of g¯
- Variance Reduction Techniques: Antithetic method
Generate path y1,y2,...,yn through sequence u1,u2,...,un from standard uniform distribution
Generate another path through (1−u1),(1−u2),...,(1−un)
Suppose the estimate of (1) using the first path is g(u) and the estimate of (1) using the second path is (1-u)
Then the combined estimator g′=12[g(u)+g(1−u)]
g′
is an unbiased estimate of
g¯
with variance