|
abstract:
Primal-dual simulation methods for constructing confidence intervals on option prices have recently been extended from Bermudan option
pricing to a quite general class of nonlinear pricing problems (see the talk by C. Bender). Such methods can be enhanced via the multilevel approach. If measured in terms of the root-mean-squared
error $\varepsilon$, the complexity of Andersen-Broadie type algorithms
for upper confidence bounds (dual problem) can be reduced to the order
$\varepsilon^{-2}$, while the plain Monte Carlo implementation of these
type of algorithms typically leads to a complexity between
$\varepsilon^{-3}$ or even $\varepsilon^{-4}$. |
