December 1st, 2004, 3:25 pm
Hi Val,We recently set up a FD convertible pricer based on the Ayache, et al papers. Unfortunately, we have found that convergence is too slow to make it practical (so we set up a trinom tree instead, but it has less flexibility). Your comment about needing 1-2 additional iterates intrigued me. We used a successive over-relaxation technique, needing hundreds of iterations to converge. What were you iterating for your 1-2 iterates?Also, I'm not at all sure our problem lies in the iterations, but possibly in the grid size, as we needed a much larger grid than 100x200 to get convergence (for the 30-year bond we were using as our test case). Do you use even step sizes in stock price (or log stock price), or some method to concentrate the steps? Any other advice?Thanks in advance