Skip to content

Skipping non-convergent GARCH run within a loop

3 messages · Anatoly Schmidt, Alexios Ghalanos

Original
Anatoly Schmidt
# 
I'm wondering if there is an option to skip a non-convergent combo of garch
input parameters within the loop upon different parameter values.
I thought something like
if(convergence(ugarchfit(GARCHspec, solver="hybrid", returns)) != "0") {
      print("NA")
      next
}
It did not give an error for the runs with 'good' parameters. But in case
of a 'bad' parameter combo, I got the following:

error in evaluating the argument 'object' in selecting a method for
function 'convergence': object 'B' not found

Thanks much, Alec
Alexios Ghalanos
# 
Hi Anatoly, try wrapping it in a ?try/catch? function. Additionally, I suggest you switch to tsgarch. rugarch is in minimal maintenance mode it?s the intention to slowly transition everyone to use the new package. With the exception of a few models/features, there is little missing from tsgarch and it is a more robust/modern GARCH package.

Alexios
Anatoly Schmidt
# 
Thank you Alexios.
Before I got your reply, I checked ChatGPT. One of its solutions was the
one I tried and reported below its failure. But another was:

try(fit <- ugarchfit(GARCHspec, solver="hybrid", returns))
    if(inherits(fit, "try-error")){
      message("  --> fit failed (error). Skipping.")
      next
    }
And it works! Surprisingly, I never get the message "fit failed (error).
Skipping".
Instead, I have the system message
Error in robustvcv(fun = f, pars = ipars[estidx, 1], nlag = nlag, hess =
fit$hessian, : object 'B' not found

(which I guess points at switching to another solver)

or
Error : $ operator is invalid for atomic vectors

but then the process continued with other parameters.

Cheers, Alec
On Wed, Dec 17, 2025 at 7:28?PM Alexios Galanos <alexios at 4dscape.com> wrote: