Gray GARCH example question (split from MSVAR topic)

[GUESMI2010]

Hi Tom,
until the new version of RATS, I applied the paper GARY 1996, I found the results with the database of gray, but with my database I found difficult.
I send you my database and the program GRAY .
THANK YOU VERY MUCH

[GUESMI2010]

Here is the program and I thank you in advance.

[TomDoan]

You posted the program for the original data set; not yours. That won't help us.

However, it's not surprising that you're having problems with the Gray model. It does not have a very well-behaved "likelihood function" as you can tell from the comments in our post. There are likely multiple modes with similar likelihood values but very different sets of parameters. I put likelihood function in quotes since it isn't the true likelihood. In order to avoid problems with path-dependence, it simplifies the lagged variance at each step. I don't know whether the source of the messy likelihood surface is that simplification, or whether it's the general problem of multiple modes shared by many switching models.

[GUESMI2010]

I thank you for the information and here is my base.
thank you very much.

[TomDoan]

That's your data set again. You said that you weren't able to get it to work with your data set. We need to see the program that didn't work correctly.

[GUESMI2010]

Hi TOM,
Here is my program, the problem is that I can not leave the graph probabilities.
Best regards

[TomDoan]

It looks as if one state is probably fine which would explain the difficulty in getting a two state model to fit.