VAR GARCH-M Model

[cczzwhy]

Hello,its me again...
I got a problem because the data get cointegration relationship,so I tried to do the VECM model,but I have no idea on how to compute the arch test,can you help me?

Code: Select all

@johmle(lags=3,det=rc,cv=cv1)
# LYEN LNKK 
**
* Define a three lag VAR with the gap being a cointegrating vector
*
system(model=vecmmodel)
variables LYEN LNKK 
lags 1 to 3
det constant ect{1}
end(system)

[TomDoan]

There's no difference. ESTIMATE the model, take the residuals and do the ARCH tests.

[cczzwhy]

Thank you very much,I 've done the arch tests ,but just got one side be rejected,can I use the GARCH-M?

[TomDoan]

Yes, but you probably don't want to include an "M" term for the series which is showing little GARCH behavior---if you do, the data underlying the M coefficient will be effectively flat and thus can't be estimated (it would be collinear with the CONSTANT). As an example, see Elder and Serletis, who only include an M effect on the oil price, not on GDP.

[cczzwhy]

Thanks for your suggestion,I also have a question about how to set the ect without log variables...

[TomDoan]

Thanks for your suggestion,I also have a question about how to set the ect without log variables...

I have no idea what you mean. You want to do mixed logs and levels in one model?

[cczzwhy]

Sorry for my question , I think maybe you mean what I mean...
I set one variable as

Code: Select all

 
set LNKK = log(NKK)-log(NKK{1})

,The other variable was calculated by myself, but the log of the variable is not needed.

[TomDoan]

What's the cointegrating relationship that you're trying to impose? As you've written that, you have log(NKK) as an I(2) process.

[cczzwhy]

So you mean I can't do the VECM?
I got two variables of macroeconomic changes ,one is set as the code attached above ,the other change is calculated by other way not logarithm,both of them are stationary but get cointigration relationship.

[TomDoan]

If they're stationary, then there is no cointegrating relationship. The cointegrating relationship has to be between I(1) variables. If you're saying that log(y) and x are cointegrated, and that dlog(y) and dx are stationary, then you want log(y) and x as the dependent variables in the SYSTEM (and in the @JOHMLE), not the changes.

[cczzwhy]

Thanks for you patience!

I need to estimate the model like this

Code: Select all

set lvix =log(vix/vix{1})
set lxt =log(xt/xt{1})
compute nlags=6

SYSTEM(MODEL=VAR1)
VARIABLES lvix lxt
LAGS 1 TO nlags

I need to confirm that do I need to test the cointegration between lvix and lxt?

[TomDoan]

Have you tested them for I(1)? I'm guessing that they don't have unit roots. If they aren't integrated, they can't be COintegrated. If anything has a possibility of being cointegrated, it would probably be log(vix) and log(xt). If log(vix) and log(xt) are, in fact, cointegrated, then VAR1 is misspecified.