full identification-FEVD problem

[KOBE24]

Dear Tom,

sorry for bothering.

I am running a Bayesian VAR with sign restrictions of order m=4 variables and I want to identify 4 shocks.
The code looks ok, and it produces IRF and FEVD as I desire.

However, I do not get why the FEVD do not sum to 100, but around 90-95 (see, for example, output-gap columns).
Does it come from the median? Should it be overcame by using the Fry-Pagan correction?
My guess is that the code I am running, provided it is correct, should achieve a full decomposition, i.e. sum 100, regardless of the Fry-Pagan correction.

Could you please give me some hints?

Thanks in advance for taking the time to answer.

Best,

KOBE

[TomDoan]

Dear Tom,

sorry for bothering.

I am running a Bayesian VAR with sign restrictions of order m=4 variables and I want to identify 4 shocks.
The code looks ok, and it produces IRF and FEVD as I desire.

However, I do not get why the FEVD do not sum to 100, but around 90-95 (see, for example, output-gap columns).
Does it come from the median? Should it be overcame by using the Fry-Pagan correction?
My guess is that the code I am running, provided it is correct, should achieve a full decomposition, i.e. sum 100, regardless of the Fry-Pagan correction.

Could you please give me some hints?

Thanks in advance for taking the time to answer.

Best,

KOBE

There's no reason for the medians to sum to 100%. The means would have to, but because of the asymmetry, the means can be outside bounds generated using percentiles.

The best advice is to not be all that concerned about error bands for FEVD's.

[KOBE24]

Got it.

Thank you Tom: I have accordingly modified the code (%avg(work) instead of frac(2)) and it is perfect now.

Your help is really appreciated!

Best,

KOBE