Hi, I'm trying to get my PhD going by writing about differing approaches to calibrating and pricing options with Levy processes.I'm currently working with the variance gamma approach and the non-parametric approach of Tankov and Cont. Can anyone speak to the relative desirability of either approach or perhaps point me in the direction of another approach. Also, any insight into what I should be looking for when comparing approaches would be very helpful as well. thanks in advance, Keith

Personally, I have never found this relative entropy business (if I know the right paper for Cont/Tankov) compelling.If you are looking at broad indices, consider the representative agent/utility function approach. (see Bates, for example,or my Wilmott mag. article "Fear of Jumps" ). To me, the rep. agent approach is under-appreciated, less ad hoc, and safer! (arbitrage-free). Here's a topic based on that (if it hasn't been done): market equil under two/multiple rep. agents with Levy processes(extending Dumas, Two Person Dynamic Equilibrium in the Capital Markets). With Levy processes, a big question for modelling is: why exchange Brownian motion (plus Poisson jumps) in favor of jump processes withinfinite intensity? The latter processes often use obscure parameters, are hard to visualize, and are much less intuitive. Given all the hassles, if you're going to go that route, *what is the compelling case to do so*? A definitive answer to this one (or rejecting the approach) is probably worth a PhD Then, there are -many- practical questions about calibration: a big one:what instruments to use besides the usual suspects (credit related?, swaps, etc.)This is a big deal for options on individual equities, which is a much less explored area, so perhaps even better to write about.The problem there is that your basic (stationary) framework should fail miserably unless you handle sensibly the knownnon-stationarities (earnings releases being the most important.). Search the forum for many discussions of "calibration", lately with tibbar or me as authors. regards,

Another neater approach could be to sort to Bayesian calibration methods as in (http://www.ingentaconnect.com/content/b ... 3/art00001)

Thanks for the advice. I'll definitely have to think about this stuff some more. d, can you post the name of the paper because that link is dead. Also, is there a baseline approach to calibration that you would compare these to.thanks again, keith