Calculating finite-horizon variance decomps?

To all:

Currently, Dynare does not calculate finite horizon variance decompositions. I did want to ask if there was a straightforward way of calculating them in Dynare.

In particular, would the following trick do the job, or, if not, can someone suggest a better method?:

Suppose we want the 8 period ahead var. decomp of GDP.

  1. Define a new variable GDPe8 = GDP(+8) (i.e. E(t)GDP(t+8))
  2. Use that to define GDPerr8 = GDP - GDPe8(-8)
    (i.e. GDP(t) - E(t-8)GDP(t))
  3. Decompose GDPerr8, which should be the forecast error on GDP given information up to t-8–I think!

Thank you all for your time and help.

Paul Corrigan

hi paul
it is easy to recover matrices T and dr_.ghu from the dynare output to use them in your own code to compute variance decompositions. T is the state transition matrix and the latter is the contemporaneous impact matrix. it is pretty straightforward to code it up, once you have saved these matrices, say in Dsgesmoother.m. just look up what Peter Ireland did in his ‘method to take models to data’ paper.