bashtage/arch

Are we using two-step approach for estimation?

barryyuberkeley opened this issue · 2 comments

barryyuberkeley commented

I may be wrong but it seems that the package is estimating the full model in two steps

  1. estimate mean model parameters (zero mean / constant mean / AR / etc..) output residuals
  2. estimate vol model parameters using the set of residuals in 1

I saw from the the summary that the degree of freedom is only nobs minus mean model parameters

If I am correct above, is there anyway to do a full MLE by finding the set of (mean parameters, vol parameters) simultaneously? It seems to be the other python package on garch is doing

Thanks very much!

bashtage commented

The estimation is all in one pass - full MLE of all parameters including mean, variance and distribution shapre.

  • 👍1
bashtage commented

Going to close as answered. Please feel free to continue to post if there this isn't clear enough.