Dear all,
I finally pushed the commits fixing the host of bugs related to trends, prefiltering, first_obs, and loglinear to https://github.com/DynareTeam/dynare/pull/852. Could you please have a look at them and provide feedback. I would be grateful if this pull request could be merged before any further large changes to the estimation interface. Integreting this was a hassle and I already had to change the code twice before due to larger updates. Thus, I would like to avoid merge conflicts as is the case for many open pull requests.
Given the by now rather large difference between the unstable and 4.4 branch, I would be very reluctant to port these fixes to 4.4. What is the plan for the next releases?
Additional points:
@Michel and @Stephane:Could you please check whether the saving of results conducted in write_smoother_results.m is as desired and discussed in http://www.dynare.org/DynareWiki/OutputVariables. The most controversial issue from my point of view is that when using the loglinear option, the results of the smoother are saved for the logged variables, i.e. I opted not to invert the logging. The reason for this is that I did not envision cases where someone actually want to estimate a model in percentage deviations but wants to compare the results to unlogged data. If you think this should be different, I can easily change it. I will also provide documentation for the changes once we have agreed the design choices are OK.
@Marco: I haven't touched any of the identification routines, but the option to estimate trend coefficients most probably has consequences for that part of the code, too.
Best,
Johannes