Bayessian Estimation


I a doing a bayessian estimation, but for some reason the plots corresponding to posterior (green and gray lines) are not shown, then everything seems correct:
estimation(lik_init=2,plot_priors = 1,bayesian_irf,datafile=Data_import,conf_sig =.95,prefilter = 1,filtered_vars,smoother,first_obs=2,mode_check,mode_compute=6,moments_varendo,mh_nblocks=5,mh_replic=200000,mh_jscale=0.5,mh_init_scale=1) y x c pih q;

Thanks in advance for your helpbol.mod (6.1 KB) lala.m (14.5 KB)Data_import.m (181 Bytes)

You did not provide the data file needed to run the codes.

bol.mod (6.1 KB) Data_import.m (181 Bytes) lala.m (13.9 KB) LALA.xlsx (17.0 KB)

Thanks for your reply,
PD: first you need to run the m file lala,

You have a nograph-option in your file.

A couple of questions more please:
is it correct to “mix” a Bayessian estimation and a calibration? or do I need to necessarily estimate all the parameters?
Also,how can I improve the Bayessian estimation, because my graphs of prior and posterior distributions make no sense at all.

Thanks in advance!

  1. No, it is standard practice to calibrate some parameter (sometimes called dogmatic priors.
  2. If results do not make sense, first check your observation equations and identification.

I am doing a Bayessian Estimation for two different countries in the period 2000-2020. But the data in 2020 is atypical, do you think is a good idea to run two different bayessian estimations, one for the period 2000-2019, and the second only for 2020?,

Myabe, a markov-switching method is better, but is it available in dynare?, where can I find an example?.

Thanks in advance

  1. Have a look at e.g. Including observable data during COVID19 period into DSGE model
    I would leave the last year out.
  2. No, Markov switching is not supported by Dynare. You want to use Junior Maih’s RISE toolbox for this.

Thanks a lot for your reply,
I had a look on the topic " Including observable data during COVID19 period into DSGE model", thus, I won´t include the data of 2020 in my bayessian estimation.
However, I found several papers that actually include these data (I attach one of them).

I am also doing a welfare analysis based on my estimation, and the ranking of monetary policy changes completely when I include 2020. So, I was thinking to make two different estimations, the first one taking into account 1966-2019, and the second one 1966-2020 and show both results. However, I am still not shure :frowning:
sustainability-13-03362-v2.pdf (1.2 MB)

I don’t think there is a strict right or wrong here. We simply don’t know whether there was indeed a problematic structural break. Views differ here. Reporting both alternatives seems like a sensible compromise.

1 Like