In the Longstaff & Schawartz article they condition on using In-The-Money (ITM) paths only for the regression. The reason for this is to obtain more accurate results and also reduce the computational burden. However, if I blindly implement this approach and the consider an American Put option that is far Out-of-The-Money (OTM) then it may very well happen that all of the paths are OTM and thus not part of the regression af they fail to satisfy the ITM-condition.
What are some goods methods to handle (far) OTM options using LSMC?
I could check if a certain percentage or number of paths are ITM before apply the condition but perhaps someone here can share some insight as of how to handle these details in practice.