Threshold Structural VAR

[dakila]

Yes. You are right (cgirf=1).

[jwojnilower]

Thank you. A couple other questions (at the moment):

1) In calculating the IRFs, does the standard deviation of a shock reflect the standard deviation of a variable during that particular regime or its standard deviation over the entire sample?

2) Is it possible to compare the information criteria of different models using the Add-in? If not, do you have any suggestions on how to do so?

Thanks again.

[dakila]

1) the particular regime
2) No. sorry I have no suggestions.

[Ovis]

Hello, I have a question. What might be the problem if I constantly receive the following error after more than 100 recursions:
"invalid (or out of range) coefficient or matrix 0"?

[dakila]

I have no idea. Could you post the data file?

[Ovis]

Thanks for the reply,

and sorry for being gone a bit. This is my excel file with US government expenditure and GDP. I do the ordering: govex, gdp and get the error message.

[dakila]

What is your threshold variable?

[nasa]

Hello,

How can I create confidence bands using monte carlo simulations for the cumulative generalized IRFs?

Many thanks,
Naser

[dakila]

It is impossible for the add-inn.

[nasa]

Do I need to expect in the near time from Eviews?

What about creating confidence bands using monte carlo simulations for linear structural VAR(shout-run)? Is that possible, please?

[Ovis]

What is your threshold variable?

It is GDP.

[dakila]

It is working. you need the time series structure.
For example:

Code: Select all

pagestruct(freq=m, start=1990) thsvar(girf=1) 2 gdp gdp @ gdp govex

[Ovis]

Can you elaborate a bit on how you made it work? It shows me "syntax error in control line". Do I need to have a variable indicating time according to which the program is guided? How exactly did you make it work with my data?

Thank you for all the help!

Edit: solved my problem. Final question: can one obtain estimates of coefficients and significance levels?

[OlleWar]

Hello fellow eviewers, this is my first forum post!

I seek some help regarding model selection in the thsVAR-case.

Since eviews returns both outputs for the upper and the lower regime, I can't wrap my head around how to compare different model specifications with the usual information criterias. Since a new threshold will be estimated if you change model specifications, the number of observations will change in each regime and the criterias cannot be directly compared.

So basically my question is; how do you go about selecting the appropriate lag length, length of the moving avarage and delay parameter? The solution may be trivial but as for now I cannot wrap my head around it.

Thank you in advance,
OlleWar

[dakila]

I think that the best way to select the length of moving average and delay parameters is try to look at the impulse response functions.
Check it is sensitive to the length of MA and delay parameter.