On the primal-dual algorithm for callable Bermudan options

This paper discusses various extensions and implementation aspects of the primal-dual algorithm of Andersen and Broadie for the pricing of Bermudan options. The main emphasis is on a generalization of the dual lower and upper bounds to the case of mixed buyer and seller exercise, along with a detailed analysis of the sharpness of the bounds. As it turns out, the method as well as the convergence analysis can even be extended to conditional exercise rights and autotrigger strategies. These theoretical results are accompanied by a detailed description of the algorithmic implementation, including a robust regression method and the choice of suitable basis functions. Detailed numerical examples show that the algorithm leads to surprisingly tight bounds even for the case of high-dimensional callable Bermudan pricing problems. Copyright Springer Science+Business Media, LLC 2013