My question is: if I prematurely terminate the iterations when the log likelihood does not change (and I have checked that the coefficients does not change either) although convergence is not achieved, can I still use the coefficients generated by the GARCH model to interpret the results?

I use the following Stata (9.2) command for the GARCH model:

• Arch delta_ln_exrate x y z, het( x y z) arch(1) garch(1)

But I’ve also tried to add on options such as “difficult” and different priming methods, but I still cannot get convergence. This is how the iterations look and the message I get:

BFGS stepping has contracted, resetting BFGS Hessian (7)

Iteration 82: log likelihood = 4273.5872 (backed up)

flat or discontinuous region encountered

numerical derivatives are approximate

Thankful for all the help I can get!