How to decompose covariance?

[akashi]

Hi, I am a new user of EViews 7.0 (student version).
I have been interested in Smets and Tsatsaronis (1997) "Why does the yield curve predict economic activity?,"
and am doing some analysis on the relationship between term spread and GDP growth, with S-VAR (GDP growth, Inflation, Short interestrate and Term spread).
I have found the way to decompose (forecast error) variance to structural shocks on view menu of VAR.
However, I am wondering how to decompose "co"variance(=predictive content) between (lagged) term spread and GDP growth.
In the paper or some other materials, I have found conceptual or mathematical explanation of the method,
but I have not seen the practical procedure (hopefully on EViews).
I have tried some silly things (adding up impulse responses manually, or many things), but everything has been failed.
It would be most helpful if I could get some suggestion on it.
Thank you very much.

akashi