Hi.I'm currently worrying about why my garch/Levenberg-Marquardt engine (forgive the specificity from here on) is sending my values for components of my alpha (cf curvature) matrix through the roof - and I wondered if there's anybody here who has had (and resovled) similar issues.briefly, Hull gives the merit function to maximise wrt the Garch parameters, omega, alpha & beta. following procedures described in Press (et al) I arrive at expressions for alpha_j,k and beta_j quite different than that given in Press (et al) for the case they consider (which wasn't worrying me untill I noticed that my expressions allow pathological behaviour...). the alpha_j,k term I arrived at includes a multiplicative term (1/v_i)*((u_i^2/v_i)-0.5)if that's clear (u_i is the return on the i'th day, v_i=sigma_i^2 and is calculated from the garch(1,1) eqn given a set of trial parameters). this isn't as elegant as I'd like, and it certainly seems to cause the alpha's to go crazy as (1/v_i) is very small (approx 10^-6) and dominates the terms in the sum.does this ring a bell with anybody?thanks in advance,Richard.

you know how sometimes when something just won't work and keeps going infinite and you're so wired on coffee and chocolate that you don't see the problem that's right in front of you??? no... well maybe it's just me problem now sorted....R.