# Longstaff Schwartz method (LSM): how to increase accuracy?

In the LSM method, I am currently (as they do in the paper) using weighted Laguerre polynomials as basis functions, about 3-5 of them.

If I wish to increase the accuracy of my model, what should I do?

1. More paths, steps, or basis functions? What will generally have a bigger impact? I could of course increase all three, but then the code may be slowed down considerably. What will give me most bangs for my buck?

2. If it's the basis functions, should I use more of the "same" basis functions (i.e. more laguerre polynomials) or should I mix the current basis functions with another set of basis functions (e.g. standard polynomials, $$X, X^2, X^3$$).

• Could you first maybe describe your current implementation as to the simulations number? Do you use one set to calibrate the exercise boundary and a second independent one to price? Do you use ITM paths for the regression? Note that LSM (when done correctly) is known to provide a low-biased estimator of the true option price. Mar 11, 2020 at 7:10