Group shocks in FE variance decomposition

Dear Dynare users

How can I group shocks for variance decomposition to see, e.g., the net effect of 2 shocks relative to 6 other shocks? (this will be particularly useful when trying to see the net effect of a foreign price shock in a SOE taking into account the deviation from law of one price.)

The manual does not seem to accommodate this as with the shock_groups(name=) and plot_shock_decomposition command for historical decomposition.

So, any suggestions on how to do this from the output results would be greatly appreciated.


Hi Hylton,

I usually extract the shock and variance decompositions from the dynare structure with the help of a separate m.file. You can then add up shocks or group them very easily. I am not aware that dynare would do this automatically for you and hence you have to do this yourself.



Hi Rob

Thanks for the reply! What I’m also interested to know is whether the FEVD can group two shocks, not aggregate the size of their contributions to variance of variable, but a net effect—like taking the (negative) correlation of the shocks to weight the contribution; let me elaborate my eg earlier:

I have a nested model with and without oil. The observed foreign oil price (shock) is naturally very large, but deviations from purchasing power of parity (law of one price) help to capture the domestic oil price impact on the economy. As a result, these two shocks generally offset each other, but because both are large they contribute majority of FEVD. Historical decomp can net the two off nicely, so I’m wondering what’s the best way to do this for FEVD…?


Dear Hylton,
i am not sure I am following. If your structural shocks are uncorrelated, the FEVD of the shocks is additive. The variance contribution of two shocks is the sum of their individual contributions.
What you seem to be suggesting is that in the historical shock decomposition, the shocks are not uncorrelated as they tend to offset each other. The FEVD based on uncorrelated shocks will not reflect this, because it is a theoretical concept. If you think this correlation is a feature of your model that you need to have, you need to allow for correlated shocks.

Hi Johannes

Yes…I think correlated shocks would be the best way to go here…thanks.

I was trying to figure out if one could somehow “re-weight” the additive contributions in the theoretical FEVD.

But, correlated shocks seems the way to go here.