I wasn't specific in the last post about pricing American options. Actually, I was wondering which Monte Carlo methods technique is the best in pricing American options. I include "high-dimensional" in this topic in order to exclude most other methods like finite difference schemes. By the best, I judge the methods on speed of convergence and accuracy without variance reduction methods. As far as I know, least-squares method (LSM) seems easiest to implement and it gives a biased-low estimator, whereas duality method provides a tight upper bound. Can one construct a biased-high estimator using LSM?

Well you can use the information you obtain from the LSM lower bound estimator as an input to the dual technique for upper bounds. See paper by Haugh and Kogan in Operations Researh. Or see Glasserman's Monte-Carlo book for a description.

Actually, there is an LSM like method from Bender for high-speed upper-bounds without nested simulation: State-of-the-art option pricing by simulationDarou

I was not aware of Bender et al's article on callable Snowballs. It is great work but most of The ideas in Bender have already been discussed elsewhere though not in context of snowballs. For example in "MULTI-FACTOR CROSS CURRENCY LIBOR MARKET MODELS: IMPLEMENTATION, CALIBRATION AND EXAMPLES" by Amin (2003) though this article does not discuss the upper bounds. It was the first article that discussed pricing of cancellable structured products directly in Longstaff and Schwartz framework. It can be downloaded from http://www. geocities.com/anan2999 .Mark joshi wrote an article with similar ideas in which he also describes how to construct an upper bound for bermudan products his article is titled "Monte Carlo Bounds for Callable Products with Non-Analytic Break Costs" and can be found on SSRN. David Skovmand also wrote an article about callable structured products titled "The Valuation of Callable Bonds with Floored CMS-spread Coupons " and this article constructs both upper and lower bound for the CMS spread products. I must say that strategy 3 with Andersen like LS method in Bender's article is certainly new in literature and a significant improvement over previous attempts to take care of the callable structured products especially for Snowball structures. I have been using it for several years and recommend it very strongly.regards,Ahsan

where can I download David Skovmand paper? thanks.

QuoteOriginally posted by: willshawwhere can I download David Skovmand paper? thanks.The Valuation of Callable Bonds with Floored CMS-spread Coupons (with Peter L. Jørgensen). Forthcoming Wilmott Magazine.