A question on (T)VAR estimation results (Fstat/ p-values/R)

[applej]

Dear Tom,
I'm estimating a TVAR model with two regimes. In one regime, I got the follow estimation results as attached. The last column(the fifth regression) is the one I'm particularly interested in. However, I find :
1.p-value of regression F is 0.28, insignificant at 15% significance level;
2.R-squares are as low as 0.1 (R bar is even lower at 0.02)
3. many insignificant coefficients exist.

I wonder can I still argue for validity of this model? If I don't want to change variables included in this VAR, any suggestions?
Especially, I think I'm depending on the 5th regression.( I'm interested in making GIRF of the 5th variable's responses to all other variables. With the insignificant F stat of the 5th regression, can the GIRF still be acceptable?)

Thank you,
J

[TomDoan]

1. Do you have different behavior for that variable in the other regime?
2. The transition points for a TVAR are based upon the overall likelihood from the five variables. You could easily have four equations that don't break at all, and one which breaks strongly.
3. That one equation looks a lot like it's white noise. Is there any strong reason to believe it should have interesting dynamics?

[applej]

1. Do you have different behavior for that variable in the other regime?
2. The transition points for a TVAR are based upon the overall likelihood from the five variables. You could easily have four equations that don't break at all, and one which breaks strongly.
3. That one equation looks a lot like it's white noise. Is there any strong reason to believe it should have interesting dynamics?

1. Yes. The estimation results are quite different in each regimes. For each equation.
2. I think there are literature supports my variable selection into the TVAR model. I picked the most mentioned relevant explanatory variables that can be determinants to the 5th dependent variable, RRG.
And by GIRF, I do observe interesting difference in each regime. I think I just wonder whether the interesting GIRF derived from the TVAR estimation can be valid(solid), given the estimated equation is not ideal.

[applej]

Dear Tom,
I got another question related to TVAR estimation. I got the following covariance\correlation matrice. covariance matrix in lower triangular.

I wonder whether this looks ok? For covariance matirx, the diagonal is almost all zeros.
And can it be possible if one pair of residuals have covariance =0, but correlation not equal to 0?

Thank you,
J

[TomDoan]

Those are only zeros because you formatted them with just two decimal places. The scales on those variables is very small (check the sample mean and standard deviation in the earlier output). If they're growth rates, multiply by 100 to make them percentages.

There's nothing inherently wrong with your results. You aren't picking one equation, you're picking five at a time with the threshold.

[applej]

Those are only zeros because you formatted them with just two decimal places. The scales on those variables is very small (check the sample mean and standard deviation in the earlier output). If they're growth rates, multiply by 100 to make them percentages.

There's nothing inherently wrong with your results. You aren't picking one equation, you're picking five at a time with the threshold.

Thank you for the quick responses. That's very insightful. I should try to rescale the variables.


Another thought: if the covariance\ correlation of each pair residuals are not 0, whether this violate "residual series are independent" assumption of VAR model?

[TomDoan]

Another thought: if the covariance\ correlation of each pair residuals are not 0, whether this violate "residual series are independent" assumption of VAR model?

What assumption is that? Serial independence is a standard assumption, but contemporaneous independence is almost never assumed.

[applej]

Another thought: if the covariance\ correlation of each pair residuals are not 0, whether this violate "residual series are independent" assumption of VAR model?

What assumption is that? Serial independence is a standard assumption, but contemporaneous independence is almost never assumed.

Indeed, thank you for strengthening it out. I was all mixed up.