Estimation with observation_trends

Dear all,
I’m trying to estimate a model on Euro data using inflation, output and interest rate.
The model relates inflation to the output gap, but I don’t detrend output to obtain an “observed” series
of the output gap. I ask dynare to estimate the output gap instead, using the equations

yn=yn(-1)+eps; // natural output process (the drift term is only in the observation_trends block)
yn=y-x;

Problem: sometimes I receive an error message that says there are problems with the diffuse inizialization
of the kalman filter for the non stationary variables, because there’s not enough information to do that.

Can anybody tell me what this is related to?
Many thanks for any reply.
Best,

Andrea

Dear Andrea, You should read chapter 5 in the book “Time Series Analysis by State Space Methods” by Durbin and Koopman (2001), the answer is around page 105.

Best,
Stéphane.

Dear Stéphane,
thank you for your reply. I’ll have a look at it.
Best,

Andrea

I found my mistake. I set an upper bound equal to one for a variable that was not
defined as unit root one. Sometimes the estimate went on the bound, so that problem
emerged.