Can I perform bayesian estimation of a non-stationary output gap filter in dynare?


The code is running but it is giving me continuous warnings:

univariate_diffuse_kalman_filter:: T does influence the rank of Pinf!

basic_4_plus_plus.mod (2.9 KB)
final_trunc.xlsx (14.7 KB)


Sorry, but I don’t understand the question. Your data is clearly still trending, which will not work.


Professor, can’t we estimate a general non-stationary state space model in dynare with diffuse filter option?
Can’t dynare (diffuse filter option) handle more than one unit root?


Also a follow up question, if the paramter that I want to a paramter that can take both postive and negative value, what should be its prior distribution?

  1. For the prior, you can e.g. use a normal distribution centered around 0.
  2. The unit roots are not the problem. The problem is the drift. You cannot have an explosive equations. You would need to use a stationary transformation.


Thank you so much professor for your response.