wilmott.com

Dear all,I'm currently pricing an American option based on convertible bond, which confused me a lot. Due to some options inclued in the convertible bond(such as option to convert to equity as well as some path-dependent options) I had to choose mento carlo but not binary tree method. In my algorithm, first I simulate the equity paths, then obtain the convertible bond paths in backward direction. After that, I try to obtain the American option price based on previous established convertible-bond paths.However, I met some troubles when calculating the American option by MAX(Bond-strike, holdvalue). In Longstaff and Schwartz's method, I need to regress the holdvalue to be holdvalue'=a*eq^2+b*eq+c on the paths that are in the money. Usually "in the money" means paths on which Bond-strike>0(if the Bond itself is the variable), but in this case equity is the variable and Bond is calculated based on simulated equities, thus someone told me I can't use Bond itself but use the expectation of Bond instead. Is that right? If I need to use the expectation of Bond which is Bond' in MAX(Bond'-strike,holdvalue'), should I regress the Bond so that Bond'=a1*eq^2+b1*eq+c1? But which paths should be included in the regression, all paths or only paths with Bond-strike>0? Could anyone give me some advices? Thanks a lot!Regards,Shunvwu