Variance Decomposition Ordering Interpretation

[finnschroeder1993]

While interpreting the variance decomposition I came across something I cannot explain myself. When I change the ordering of the variables in the VAR the Variance Decomposition of the variable that is listed first stays the same (Although It is now a different variable). Shouldn't the Variance Decomposition move with the variable. Is something wrong with my model? I attached the Variance Decomposition Output as well as the IRF for two different ordering of the variable. Order 1 DITEFA DEUROSTOXXB DVSTOXX DIR Order 2 DEUROSTOXXB DITEFA DVSTOXX DIR

[TomDoan]

The effect on the impact FEVD isn't a surprise---it's matrix algebra. (Basically, it's because the correlation of A with B is the same as the correlation of B with A). Take any VAR and flip the order of the first two variables and the percentages of 1 vs 2 for step 0 will always be the same. What's calling your attention to this is that the two series are basically (contemporaneously correlated) white noise, which is why the responses are pretty much flat zero from step 1 on. Thus, there is nothing really to look at beyond the impact.

[finnschroeder1993]

Thank you very much for the prompt and detailed answer. Do you have any suggestions on changes to the VAR Setup? Would you suggest running different models?

[TomDoan]

Your first three series have rather highly correlated residuals, which suggests that you should be looking at a structural model, rather than relying on orderings.

[finnschroeder1993]

Thank you Tom.