Hello guys, I am trying to replicate a paper that does a bayesian estimation of a DSGE model with bubbles. The model is kinda big and it estimates 19 parameters using bayesian methods and it has 2,000,000 draws of the MH algorithm. I need the posterior mode and mean and the whole estimation process takes about 17 hours. I wonder if there is a way to save time in doing the estimation. I tried the following:

first I calculate the whole MH using the following command

it works fine and it gives the mean, the mode, and all the info I need

, however, since I did not get the results that I wanted I tried changing the drop and the jump scale values, but using the MH draws that I already have by using the following command,

and I got the similar results that shows in [Oo_.posterior_mean is missing)

when I use mh_replic=0 in combination with load_mh_file I just get the new posterior mod, but not the posterior mean.

is there a way around this?

Best Mario

Have you tried the recent snapshot? What exactly is the problem?

Based on my current understanding, the answer is no. You are loading a mode_file but donâ€™t run an MCMC chain. What is changing the percent of draws dropped at the beginning and the dispersion of the proposal density supposed to do when you are not running a chain?

If you just want to drop more draws at the beginning, you could do that manually by loading the mh-files.

Sorry, maybe I was not clear (tends to happen). I need to do a Bayesian estimation of a DSGE model. I need to do 2,000,000 draws (that is what the paper that I am replicating says). I have estimated the model once, can I use those 2MM draws that I get for the first estimation to get another estimate if, for example, I see that the acceptation rate is not good?

As I tried to state before I use the mh_replic=0 because I already have all the draws that I need, but I still need to get the mean and mode.

I am sorry if I am not clear, I am new using dynare and doing bayesian estimation

thanks