Time varying SVAR

[dakila]

Training sample is used for determining prior parameters of Time varying SVAR. There is no test for determining the number. It should be at least 30-40 depending on VAR size.
Because it uses training sample to estimate OLS VAR. In other words, training sample is not used for actual TVSVAR model. If you want use full sample, then you can use uninformative priors (flat) theoretically. But these option is not available for this add-in for the reason of convergence problem.
Lastly, you need to upgrade your Eviews to version 9.

[samiri]

Hi, I want to estimate a TVP SVAR with three variables, taxes g GDP. the data are frome 1998Q1 to 2014Q4. And I put in the training smple 40,30,60. And every time I get the following message "the GM01 size is not as the same lenthe simple". Please could somone help me .
and thank you.

[dakila]

What is the version you using? 7 or 8?
Can you run tvsvar_ex.prg (example file)?

[samiri]

Im using Eviews 9 and i can run the Primiceri example. and the variables are in log and first difference, and I tried by eliminating the NAs, and i get the same erreur "GM01 is not as the same lenthe simple"

[dakila]

Could you post the file?

[ali_Economist]

Hi, Thank Dakila for help on my first question, I have downloaded the add-in, that looks ok, however when following document instructions, I put the details in it gives message "Near singualr Matrix". I don,t get any impulse response but some series and matrixes appear when do not make much sense to me.Please help.

Regards

[samiri]

the file contains three variables g taxes and gdp all in log and first difference

tvp var .wf1

the file contains the variables in log and first difference

(16.72 KiB) Downloaded 629 times

[dakila]

There is a bug. I will try to fix it. However you can use the following code to fix it:

Code: Select all

pagestruct(start=1998q2)

[dakila]

Hi, Thank Dakila for help on my first question, I have downloaded the add-in, that looks ok, however when following document instructions, I put the details in it gives message "Near singualr Matrix". I don,t get any impulse response but some series and matrixes appear when do not make much sense to me.Please help.

Regards

1. Try to transform your variables. For example, to annual growth rate
2. Try to change the default prior parameters.

[samiri]

Ok I will try by resizing the simple. thank u very much dakila

[ali_Economist]

Hi, I want to estimate the TVSVAR model on the attached data file, when I try using the instruction in the add-in document it does not give me IRF output or even the estimation results, I wonder if somebody can help me please.

[dakila]

The following code is working for me.

Code: Select all

svector dtsel=@wsplit("2000m12 2005m12")
tvsvar 2 40 dtsel lnoilprice @ lnftse lnoilprice lnoilsector

[ali_Economist]

Thank you very much for reply, but it gives message that "DTSEL" is not defined. Please help.

[nadybe]

Hello Davaa,

I used the data someone post earlier (the primisceri data) to understand the steps on applying the TVPVAR but i have some questions, is it possible to obtain the FEVD (variance decomposition)? it seems that your program doesn't return it.

Best regards

[dakila]

No, it is nonsense to estimate FEVD