About the last question, I mean when I run my .mod I get options_.nobs = 32 in unstable version, which is size of my data,
but in 4.4.3 and previous versions I get options_.nobs = 28.

By other side, when I try to use dsge_var_likelihood function running this:

I get size(PHI_tilde,1) = 20. But It should be 10 (=5x2) because I have five observable variables and I set

options_.varlag = 2;

What would be the rigth form to loop over dsge_prior_weight option and over sample size to get PHI_tilde and SIGMA_u_tilde. I loop over dsge_prior_weight in a .mod and I loop over sample size in a .m file. But I see when I loop in a .mod (over dsge_prior_weight) dynare estimates twice, it makes slow this method, because in each iteration I run my .mod and after I run dynare_estimation(var_list_). I attach my files. question.rar (17.4 KB)

1- That is a matter of the changed internal treatment. In 4.4.3 nobs deducted the options_.first_obs. But that should not affect the results.
2. You are setting the wrong option for the lag number. It should be

Thus, you are still at the default of 4.
3. You should be able to change the prior weight within a loop via