EViews.com

Hi, I have a question regarding to asymmetry effect in GJR-GARCH models, maybe my question is somehow irrelevant to Eviews forum but I will be thankfull if someone reply me. I am running GJR-GARCH model to analyze asymmetry and leverage effect in some commodity markets. I tried two models, 1. simple GJR model 2. GJR model with including oil shocks as dummy variables to the variance equation.
The problem is that for some commodities the simple model (without energy shock dommies) doesn't show asymmetry effect but when I include energy shocks dummies to the variance equation of the same model then asymmetry term becomes statistically significant.
Can any one help me that how I can interpret it according to the econometric properties of asymmetry GARCH models?
Thanks

Hard to say without seeing the actual data.

Hard to say without seeing the actual data.

Thanks for your reply trabadur. Here I attach the Eviews file related to Aluminium return, in the workfile: 1. Aluminium_normal_1 is the simple GJR with normal distribution, 2. aluminium-GJR_normal_oil1 is GJR with including oil dummies, 3. aluminium_studeny_1 is another GJR with student-t diestribution, 4. aluminium_studeny_oil1 is model 3 but with including oil dummies, 5. final_normal_1 is GJR with normal distribution but with using outlier corrected data, 6. final_normal_oil1 is the same as model 5 but with including oil dummies, 7. final_student_1 is GJR with using outlier corrected data and student-t distribution and 8.final_student_oil1 is the same as model 7 but with oil dummies.
as you may see when I correct the data from outliers (I used doornik approach that you proposed me in another topic), then there is not asymmetry effect, but original data shows asymmetry effect, however when I add oil shocks dummies to the model that use outlier corrected data then asymmetry term become statistically significant again, can you tell me how can I interpret both of these changings? I mean why when data is corrected from the outliers then asymmetry term become statistically insignificant? and why when I add dummy variables it get significant again?

Im doing this estimations for all metals and each of them shows different behaviour, for example for copper is exactly opposite behaviour with aluminium and models with oil shocks dummies dont have asymmetry effect but without that then asymmetry terms get statistically significant. Im wondering if there is any explanation at least for one of them?

Thank you in advance

Asymmetry component measures the impact of negative shocks, if any. In this particular model, outliers seem to have been the very source of this leverage effect. Cleaning them solves the problem of asymmetry.

You define oil shocks as extended pulse variables. I am not sure if it is appropriate, since the variance equation is already conditional on its past values. These windows change the dynamics of your variance equation and may induce asymmetry. You can either consider modeling these shocks as single pulses or put them in the mean equation instead.

Asymmetry component measures the impact of negative shocks, if any. In this particular model, outliers seem to have been the very source of this leverage effect. Cleaning them solves the problem of asymmetry.

You define oil shocks as extended pulse variables. I am not sure if it is appropriate, since the variance equation is already conditional on its past values. These windows change the dynamics of your variance equation and may induce asymmetry. You can either consider modeling these shocks as single pulses or put them in the mean equation instead.

Thanks Trabadur, I tried all metals (10 metals) with including oil dummies to mean equation instead of variance equation and the results are exactly the same as before, just the coefficients are slightly different but the statistical significance of the asymmetry terms remain the same as before, means that without oil shock dummies no asymmetry but with oil dummies there is asymmetry for most of the metals. this happen
in this case is there any explanation?
Thanks

I have already explained the possible reasons of this behavior along with two plausible remedies. The instability of your model may also be due to the use of higher order GARCH models. Try more parsimonious specifications (e.g. p=q=1).

I have already explained the possible reasons of this behavior along with two plausible remedies. The instability of your model may also be due to the use of higher order GARCH models. Try more parsimonious specifications (e.g. p=q=1).

when I use GARCH(1,1) model, the ARCH effect remains in the residuals thats why I used GARCH(2,2). you mean different remedies together that I used: 1. correcting outliers 2. student-t distribution and 3. including dummy variables, can make confusion, yes its correct , this result is obtained mostly in the models that I used various methods together to capture the fat tails, but in a few cases I see this kind of results even with some simplest models: original data with normal distribution. I think there is not a logic way of explanation for that.
Thanks for your help