Pricing Bermudan options using regression trees/random forests

The value of an American option is the maximized discounted cash flows from option. At each time step, one needs to compare immediate exercise with continuation and decide as soon strictly greater than value. We can formulate this problem a dynamic programming equation, where main difficulty comes computation conditional expectations representing values at step. In (Longstaff Schwartz, 2001), these were estimated using regressions on finite-dimensional vector space (typically polynomial basis). paper, we follow same algorithm; only are Regression trees or Random forests. discuss convergence LS algorithm when standard least squares regression replaced trees. Finally, expose some numerical results random forest gives excellent in high dimensions.