Hey Wilmotters!I have a question about Jump Diffusion models. I have recently started to play around with them. I am looking at models with a log-normal jump size component (mean reverting). I am fitting these to spread series.I am estimating the parameters using a package called "R" - using a constrained BFGS algorithm. I am forming a likelihood function using the Ball and Torous density. I fitted these models to weekly credit card series...and the simulations looked quite good. However, when fitting the model to a particular series I get rubbish simulations. The estimated probability of jump is very high. I am curious to understand the issues underlying the estimation. I understand the likelihood function can have associated degenerecies.All comments would be welcome.Rgds,MC.

You shoud have a look on bayesian estimation for mixtures and MCMC. Let me know if you are interested.

Hi,Thanks for your response. I have read quite a few papers on MCMC estimation (by Neil Shephard,Eraker, etc)...some time ago though!However, I don't really wish to spend time designing an MCMC algorithm for a jump diffusion model.Unless, you have a tried and tested one for me...I am using MLE, it seems the R algorithm has difficulty in estimating the alpha parameter (mean reversion) - the jump probability also seems to be quite high.Rgds,

see your private inbox

ne1?