DSGE irf's from Bayesian estimation

[tomnilson40]

Hello,

I am getting the DSGE irf's after estimating the model using the Bayesian method (mode_compute=6) but the problem is that eventhough the initial values are set to be equal to zero but it seems that the irf's do not start from zero but they all get back to zero!

Does it calculate a new initial value for each variable after estimation and before drawing the irf graphs?! if it does, but why they all get back to zero?!!

If I simulate the model (stoch_simul ( ...) )they all start from zero and get back to zero.


I would appreciate if anyone could help me on this.

Cheers,
Tom

[jpfeifer]

If you have consumption and there is a shock, it will always move immediately. Thus, the IRFs from estimation strike me as normal. I am rather puzzled by the stoch_simul-IRFs not moving at the beginning. This could only happen if ALL your variables are predetermined.

[tomnilson40]

Dear jpfeifer,

Thank you very much for your prompt reply as always.

Yes, I have consumption in the model and a couple of jumper variables. But let me ask two more detailed and maybe simpler questions to clarify my previous message:

1) At time zero in IRFs from estimation, should the value be equal to the initial value determined in the model (in my case equal to zero) or the shock affects the variable exactly in time zero and thus changes the value to something else other than the determined initial value?

2) Let say that I estimate the model after making sure that simulation works. If I replace the related parameters with the estimated parameters and run the stochastic simulation (“stoch_simul”), should I get the same results as the dsge IRFs after estimation (using “bayesian_irf”)?

Thank you very much again for your time.
Cheers

[jpfeifer]

1) It's the latter. The IRFs are not that at the first plotted period we are in steady state and then the shock hits. Rather, the first period with the shock is plotted.
2) No. They are not exactly the same. The stoch_simul IRF will be at the mean parameters while the Bayesian IRFs are mean IRFs integrated over the posterior.

[tomnilson40]

Dear jpfeifer,

I do appreciate your responses to my questions. It helped me a lot.

Have a great day!

Cheers,
Tom