Closed xuJ14 closed 1 year ago
What happens when you simulate a long time series and then estimate parameters? They should be close. If they aren't, this there is an issue with your log-likelihood.
As for the constraints, they are very loose and should not normally bind. They are there to help the minimizer avoid regions of numerical instability.
One thing I notice is that you are failing to initialize the backcast value for lnsigma[0]
. Your starting values are also quite bad since the intercept should be of the order (1-beta)*log(var).
I'm going to close as this isn't really an issue in the package. Feel free to continue to post relevant information.
I wrote my own EGARCH model via python. When comparing the result from arch_model, they are totally different, both values of parameters and loglikelihood. I Wonder:
So,
I find out that in package arch_model, the authors seem adding some conditions/boundries to the parameters, which disagrees with the principle of EGARCH.
The result is so different, which really makes me confused. Am I wrong?
Sorry for the massive math script.