Estimation with missing observations using Dynare 4.5.3

Estimation
1

Dear Professor Pfeifer,

Dynare 4.5.3 Bayesian dsge model estimation fails (“Likelihood is not a number (NaN) or a complex number”) if there are missing observation(s). For instance, the ls2003 demo can’t be estimated using v. 4.5.3 if only one arbitrarily selected observation is deleted, while the model with the same missing observation is easily estimated with v. 4.4.3. I would appreciate any guidance on how to do estimation with missing observations with v.4.5.3.

Thank you very much in advance.

2

Thanks for reporting this. The problem is the combination of the prefilter option and missing observations. You can fix this issue by changing two lines of code in the Dynare file matlab/utilities/general/demean.m as shown in https://github.com/DynareTeam/dynare/pull/1583/commits/92549e48b12c60165693f9ea1289825e88b35cdd

3

Following your instructions, the problem has been solved. Thank you very much.