MCMC matrix variance covariance

Estimation ML/Bayesian estimation
1

Dear all,

I’m trying to save the matrix variance covariance from my estimation to use in a second estimation with the function mcmc_jumping_covariance.

I’m using the function mcmc_jumping_covariance = hessian, for the first estimation and for the next one I would like to use the matrixes obtained in the first estimation and drop half of them.

How can I?

  1. Save the matrix obtained in the first estimation;
  2. Drop half of the samples regarding the matrixes obtained in the first estimation;

Also, I would like to import the matrix for the second estimation using the code:
mcmc_jumping_covariance = FILENAME.

Thanks in advance,

2

What do you mean with

?
For each MCMC, there is only one proposal density covariance.
Also, you may find tests/estimation/fs2000_MCMC_jumping_covariance.mod · master · Dynare / dynare · GitLab useful.

3

Yes, I was wrong.
Actually, what I want is to use the previous matrix variance covariance that I estimated using MCMC method in the second estimation and so on.
In each simulation I will discard a number of parameters vectors to use only a part of them for the next simulation.
So, what I would like to know is how can I save a matrix var covar and load this matrix in my next estimation? And how can I discard a number of parameters to use only a part of them in my next simulation?

4

You mean you want to compute the covariance matrix between MCMC draws after a burnin of 50%?

5

Yes, I would like to do a first estimation of a matrix var cov = identity with the entire sample. For the second estimation, I would like to use the matrix var cov that I estimated in the first estimation and also burn 50% of the parameters vectors obtained.
I will do this process 4 times, and each time I will burn 50% of the parameters vectors and load the previous matrix var cov obtained in the previous estimation.