Hi,
I want to make a consult on Instantaneous Causality. First, Is it possible to test it for a VARX model?, or do I need to first include the exogenous variable in the VAR?. If it is possible, How can I do it in RATS?. I've already searched in the User Guide and some other resources. Is it easy to programming it?.
Thanks,
Instantaneous Causality in VARX
Re: Instantaneous Causality in VARX
What does this being a VAR-X mean to this question? "Instantaneous causality" was a concept that was introduced 40 years ago and shoved aside as it really didn't mean anything. The innovations are almost always contemporaneously correlated (the instantaneous causality test is for the correlations being non-zero), so the main question became, how do you explain them (through recursive orderings or structural VAR's)?
Re: Instantaneous Causality in VARX
Thank you Tom.
What I mean with VAR-X is a multivariate model of Y1t and Y2t that does not only include an auto-regressive part (Y1t-1,Y2t-1,...,Y1t-p,Y2t-p) but also exogenous variables (X1t). I've already used the Wald test for Granger Causality. But It would also be interesting to review the instantaneous causality, even if the results are highly predictable. The thing is that it is not possible to run the correlation test since the residuals (U1t, U2t) are on Y1t and Y2t not on X1t. Any idea on how to test it?. I've been thinking of doing the Wald test testing X1t+1. Do you think is a good approach?. Some useful information of X1t is that it changes with less frequency than Y1t and Y2t so it wasn't interesting to use their lags.
Esteban.
What I mean with VAR-X is a multivariate model of Y1t and Y2t that does not only include an auto-regressive part (Y1t-1,Y2t-1,...,Y1t-p,Y2t-p) but also exogenous variables (X1t). I've already used the Wald test for Granger Causality. But It would also be interesting to review the instantaneous causality, even if the results are highly predictable. The thing is that it is not possible to run the correlation test since the residuals (U1t, U2t) are on Y1t and Y2t not on X1t. Any idea on how to test it?. I've been thinking of doing the Wald test testing X1t+1. Do you think is a good approach?. Some useful information of X1t is that it changes with less frequency than Y1t and Y2t so it wasn't interesting to use their lags.
Esteban.
Re: Instantaneous Causality in VARX
I'm still confused. Are you trying to run the test on the X, not on the Y's? What is the "Granger causality" test that you've done?
Re: Instantaneous Causality in VARX
For the "Granger Causality" I've used the wizards that RATS offers, so I restricted the parameter of X to be zero. According to the user guide, this procedure implements a Wald test. Yes, exactly I want to test it (the instantaneous causality) on the X. Thanks again for your guide.
Re: Instantaneous Causality in VARX
That's not a test for Granger causality, it's a test for whether X is a significant explanatory variable. And wasn't that already a test on the current value?
Re: Instantaneous Causality in VARX
Hi,
Thank you, you are correct, if the variable is included in T the instantaneous causality could be analyzed. I will think about it again. About the wizard option from vecmgarch.rpf, I understood that using the Wizard to excluded the variables is a way of testing causality, Isn't it?. If the option is yes, Could you please attach a resource for further reading on why it works as a test for causality?
Your answers have been very useful.
Thank you, you are correct, if the variable is included in T the instantaneous causality could be analyzed. I will think about it again. About the wizard option from vecmgarch.rpf, I understood that using the Wizard to excluded the variables is a way of testing causality, Isn't it?. If the option is yes, Could you please attach a resource for further reading on why it works as a test for causality?
Your answers have been very useful.
Re: Instantaneous Causality in VARX
You may be just applying terminology incorrectly. Granger causality is a test for a specific type of dynamic relationship. "Instantaneous causality" is a term that you should just forget you ever heard. Granger causality is linked to exogeneity (or lack of exogeneity). "Instantaneous causality" is linked to nothing of any real use.
It sounds like you are embedding this in a GARCH model---is that correct? If you are putting the X variable in contemporaneously, is it really appropriate to do that (i.e. is it really exogenous or at least predetermined with respect to the Y variables?) If not, you can really make a mess out of a GARCH model which is assuming that the residuals are one-step-out prediction errors.
It sounds like you are embedding this in a GARCH model---is that correct? If you are putting the X variable in contemporaneously, is it really appropriate to do that (i.e. is it really exogenous or at least predetermined with respect to the Y variables?) If not, you can really make a mess out of a GARCH model which is assuming that the residuals are one-step-out prediction errors.