EViews.com

alvarezcc wrote:Hi! I have a question regarding the selection of the moving average order and the delay parameter. What is the best way to select this options? For example, if using a moving average order of 2 I obtain a lower value for the logdet(sigma) when using ma=3, does this means that I should select the ma order of 2? This same logic applies for the delay parameter?

Thanks in advance

I think that the best way to select the length of moving average is try to look at the impulse response functions. Check it is sensitive to the length of MA. The logdet(sigma) does not matter for choosing the length of MA. It matters for selecting the threshold level conditional on some parameters of the model.

I cannot find the tarcoint add-in in the eviews add in. Where can I get it? In case if I want to study the long run effects.
As you know if variables are co-integrated, econometricians suggest to use VECM than VAR. It means that if someone uses VAR instead of VECM, he may lose the long run information. So, you mean that I can use the VAR if I am interested only to study the short run effects even though the variables are co-integrated? Cannot be say the model is misspecified? From threshold model aspect, threshold SVAR to identify the shocks to study the short run effects. SVEC models study both short and long run effects, though do not know about this from threshold model aspect.Any suggestions?

What about a structural break or other dummies not included in thsvar add in, could it pose a problem in the model result?

you can download it from http://www.eviews.com/Addins/addins.shtml or go Eviews Add-ins menu then Download Add-ins.
Theoretically VECM sounds nice. However in practice it does not work in my opinion. If you care about the long run relationship then you can use Threshold SVAR model in levels.

Thanks. What about the structural break and other dummies not included in the thsvar add in? Could they pose a problem to the model result? Many thanks.

Threshold SVAR model itself is trying to capture the nonlinearities or structural break. It is one type of regime switching model.

Hi! Thanks again for your kind reply. I have another question. The simrep option allows to choose the number of bootstrap replications for Hansen's test. However, no matter what number I put in the instruction, it always perform the same number of replications (recursive iterations). I have tried several numbers (200 or 1000) but it always perform 71 replications (at least that's the number Eviews shows before starting to compute the IRF's). Do you have an explanation for this? Or maybe I'm just interpreting something wrong.

Thanks!!

Yes, simrep option allows to select the number of replications for Hansen's test. However the recursive iterations is different animal. The thsvar add-in estimates Threshold SVAR model recursively for the every threshold value. That's is why you performs 71 recursive estimations.

Hi,
How can I generate historical decomposition for this regime switching model? If this is not available on eviews ,please try to include. Thanks

It is not available. I will try.

Hi again! I just want ask how many replications are performed when computing the GIRF's. These are not necesarily the same amount of replications necesary for Hansen's test right?

Thanks!!

number of replication = number of variables * number of bootstraps * number of regime (upper or lower) sample

Hi,

If I want to estimate the endogenous variables in levels and include a time trend so as to take care of the long run information,how can I do it in this ThSVAR add in,please?

Thanks,

Thanks for the add-in. I am trying to estimate a threshold SVAR. I've gone through the process of estimating a SVAR including estimating the coefficients in A and B. Problem is I am stuck thereafter. I keep on getting an error message "na found in matrix". It seems there's something I am not doing right. How do I move from a SVAR to a threshold SVAR??????? I am using eviews 8.

upgrade to 9

I've upgraded to 9 but i still get the same message.