Re: Reg-TVAR models
Matthieu Stigler
Please actually ask your questions to the ts...@googlegroups.com so that
other might see your questions and maybe answer
Le 19/07/2011 08:43, devi kamatchi a �crit :
>
> Respected Sir,
>
> Sir, i have the following questions. Please reply me,
>
>
> 1. Is there any indicator or tools for model order selection for TVAR
> models (such as autocorrelation in univariate model building, since
> the lags in acf and pacf will indicate the order)?
not that I know for the multivariate case. There is a paper of Eithreim
and Terasvirta where they study the behaviour of conventional
misspecification tests in the stera/star univariate framework. For the
multivariate case, you might want to use/adapt the test from urca/vars R
packages
>
> 2.I know about AIC (minimum aic will lead to best fit) but it may lead
> to overfit, so i need some other tools for model order selection.
You will find in Franses and van Dijk (2000) book a discussion on order
selection.
>
> 3. In TVAR models, the thershold variable is only derived from the
> first column of the multivariate time series. what is the reason and
> how the threshold is estimated?
this can be changed with argument:
mTh: combination of variables with same lag order for the
transition variable. Either a single value (indicating which
variable to take) or a combination
For the model, references are given in the help file as well as in the
vignette. It is simply CLS.
Best
>
> Please reply me sir, I have studied many literatures but didn't get
> the answer.
>
> Thank you very much in advance.
>
> --
> S.DEVI KAMATCHI
> IIT MADRAS
>