wilmott.com

QuoteFigures are actually those and there are some cases, especially those with larger sizes in which scaling is superlinear. I conjecture that the explanation must stay in the fact that within Gauss OpenMP and pthreads are managed in a way which makes them more effective as the size of the problem grows. With respect to PDF/FDM, maybe another explanation of this superlinear or linear scalability stays in the fact that in multivariate lattices the nest of loops -- i.e. nodes in the discretized grid -- shrinks as the algorithm approaches the final result. I should try my codes on larger machines for just to verify whether this scalability keeps the same linear pace or it deteriorates as the number of cores goes up.The difference in speed is clear in retrospect; PDE calculates the solutions at all mesh points while backward induction is on an ever-decreasing lattice. But you only get the price at one point.A decent dynamic scheduler (as in OpemMP) helps.QuoteMonte Carlo methods for American Options, such as LSMC or Quantization approaches -- those I never managed to replicate -- are not so stable, as a matter of fact many articles use to benchmark LSMC results with multivariate lattices ones.Fair enough. You could test just plain options on MC to determine the effort and speedup compared to your lattice method.QuoteI could post a 2 factor example which could work on a 4/8 commodity machine, but only when the paper gets published in a decent journal.What's the rationale? It takes years to publish and in the meantime you need feedback.

QuoteWhat's the rationale? It takes years to publish and in the meantime you need feedback.this is the second paper in which I am caught in a covetous interest in my codes while my essay becomes second order.It seems to be the programmer's curse. People are not interested in what you write but in what you program.As I wrote previously, I would be eager to share my Gauss codes in some repository but I would like also to get out of my work the only reward most researchers can get out of their effort, that is a publication. Swapping publication time with posting time would expose me to the dangers that somebody else would take advantage of my programming effort publishing instead of me.After sincerely explaining my rational fear, I would be very grateful to you if you could let me know which journal and/or conferences and seminars my paper can get some feedback from, without being exposed to unconditional codes disclosure.Aren't there some Willmott's seminars ? thanks to all, and happy new year 2014.

One option is SSRN for working articles.

Thanks for your suggestion.My paper is already thereclick here to be redirected to http://papers.ssrn.com/sol3/papers.cfm? ... id=2365005

that may be a good idea. Thanks.

QuoteOriginally posted by: CuchulainnOne option is SSRN for working articles.Of course. Lapse of memory on my part

Maybe I asked this question before, but it is possble/useful to create lattice once and price different product off it?

Yes it is possible. As a matter of fact, in my codes I use to price simultaneously 8 options, namely Call on max, Put on the max, Call on the min and Put on the min, in both European and American or Bermudan Styles.Moreover, my parallelization approach could be deployed also for impulse control Bellman's Dynamic Programming,as you may find out in a previous paper of mine, Please click here to be redirected to "Assessing Least Squares Monte Carlo for the Kulatilaka Trigeorgis General Real Options Pricing Model"There, I benchmark LSMC with multivariate lattices. When I wrote those codes, multi threaded syntax in Gauss was at its first versions and I did not use it.Possibly, in 2014 even those codes can be modified to accommodate parallel programming approach reported in Please click here to be redirected to "Going Parallel over the Rainbow" As a matter of fact, rainbow options are my sandbox to test convergence of multivariate algorithms and models. Moreover, pricing 8 for each code execution allows me to test the capabilities of my hardware as if each of those eight option contracts were an operating mode in the Kulatilaka Trigeorgis General Real Options Pricing Model.