Additive and multiplicative duals for American option pricing (Q2463707)

It is known that the optimal stopping problem admits dual formulations, where maximization over stopping times is replaced by minimization over martingales. Such dual formulations can be additive or multiplicative depending on what decomposition of positive supermartingale is considered. The objective of this paper is to compare and further investigate these additive and multiplicative duals to the American option pricing problem. The authors consider only the ``Bermudan'' case of a finite number of exercise opportunities. Using new transformations of martingales, they show that any additive dual can be improved by multiplicative dual and vice versa. This iterative improvement converges to the optimal value function. Also, growth in the variance of estimates of value function based on two duals is compared as the time horizon grows. The conditions are given under which the variance using the multiplicative dual grows at least exponentially and the variance using the additive method grows at most quadratically.