How much faster is Dynare with SilverBullet?

[SamuelHurtado]

Hi

I recently posted, in the "contributions" forum, a tool to accelerate the estimation of big models using Dynare:
viewtopic.php?f=2&t=1609&p=2745#p2745 *

For the model I'm estimating, it makes Dynare run 6 times faster, while giving exactly the same results
with another version of the same model it run 10 times faster, again with numerically identical results
(you have to set seeds for the random number generators before launching Dynare to get identical results)

Anyway, I'm not sure if this is a general result or if it only works with our particular model, so please, give it a try and let us know: is it working for your model too?
(the first vote is mine; please try a reasonable number of M-H iterations: with 100 it can't do its magic properly, as most of the time is spent analyzing the results, and that's not going to be any faster)

NB:* new url

[iacoviel]

I have tried on my version of Dynare 3. For a model with 72 equations, it is 1.5x faster. However, the same model written on Dynare 4 - without the silver bullet - runs 6x faster (4 times faster than Dynare 3 with silver bullet).

Matteo

[SamuelHurtado]

thanks a lot for the info
it means this is just a workaround until version 4 is no longer "meant for testing"

still, it is working like magic for us... it is surprising to see how I managed to solve my problem without solving anybody else's...

[tomiw]

similar results here

it's about 1.5-2 times faster for the estimation of a smets-wouters-type model.
but 5-6 times faster with dynare 4.

but a lot stuff haven't been implemented in dynare 4 yet (like forecasts). you still solve my problems though.

thanks and best, tomi