Forecast uncertainty - VAR vs BVAR

Topics: Kalman Filtering, Models, VAR, SVAR, FAVAR
May 9, 2014 at 3:33 PM
Hi folks, I understand that forecasts under BVAR include parameter uncertainty while VAR does not. Is it appropriate to say that the uncertainty around the forecasts of DSGE models estimated with Bayesian techniques also include parameter uncertainty?

Thanks!
Coordinator
May 9, 2014 at 4:13 PM
Hi Hector,

if you're talking about the std errors returned by forecast(...) or jforecast(...), these always only include future shock uncertainty (based, in addition, on a linear approximate solution), be it VARs, BVARs, or DSGEs.

In order to assess parameter uncertainty, you need to set up a procedure where you draw from the parameter distribution a large number of times, run a forecast for each draw, store the forecast distribution (based on shock uncertainty), and then aggregate all the forecast distributions across all parameter draws.

Hope this help,
Jaromir
Marked as answer by jaromirbenes on 5/9/2014 at 9:14 AM
May 9, 2014 at 10:39 PM
That makes sense. Thanks for your reply.

Hector