Bootstrap VAR [Simple/Bayesian VAR]

[bco]

Dear all,

I'm bootstrapping a simple VAR (with 4 variables; that was converted to a model "MODEL_F") - in order to generate a few thousand forecast over a sample (see code below):

MODEL_F.stochastic(i=b, d=f, v=t, s={%simulation_period}, r={%no_simulations}, c=f, p=BS)
MODEL_F.solveopt(s=s, o=b, struct=t, w=t, v=t)
MODEL_F.solve

My question is how the bootstrapping procedure works exactly, in the sense that the forecast probably converges to the mean over the long run - but there are resampled shocks added on top of the forecasted values (over the forecast)? In each forecast, the sample is reshuffled and the VAR re-estimated (before generating the forecast)?

I've searched for the answer in the Eviews (8.1) help, the manual(s) provided (pdf(s)) and in the Jean Louis Brillet's manual/guide (http://www.eviews.com/StructModel/structmodel.html).
However, I could not find an answer.
Could you help?

[EViews Gareth]

The VAR is not reestimated. Random (normal) numbers are added to each equation.

[bco]

Dear Gareth,

the option that I choose in the model_name.stochastic(options) is i=b (i=arg (default=“n”) Innovation generation: “n” (normal random number) or “b” (bootstrap)).
Therefore, how and in which way the bootstrap forecast is different [than the normal random number option]? Is there any reference to the exact equation/explanation of this option(s) in the model (solve, stochastic, solveopt)?

Thank you!

[EViews Gareth]

The existing residuals are bootstrapped rather than generated.

http://www.eviews.com/help/helpintro.ht ... el.html%23