On the Suboptimality of Single-Factor Exercise Strategies for Bermudan Swaptions

In this paper we examine the cost of using recalibrated single-factor <p> models to determine the exercise strategy for Bermudan swaptions in a <p> multi-factor world. We demonstrate that single-factor exercise strategies <p> applied in a multi-factor world only give rise to economically insignificant <p> losses. Furthermore, we find that the conditional model risk as defined <p> in Longstaff, Santa-Clara & Schwartz (2001), is statistically insignificant <p> given the number of observations. Additional tests using the Primal-Dual <p> algorithm of Andersen & Broadie (2001) indicate that losses found in <p> Longstaff et al. (2001) cannot as claimed be ascribed to the number of <p> factors. Finally we find that for valuation of Bermudan swaptions with <p> long exercise periods, the simple approach proposed in Andersen (2000) <p> is outperformed by the Least Square Monte Carlo method of Longstaff & <p> Schwartz (2001) and, surprisingly, also by the exercise strategies from the <p> single-factor models.