I am also finding that posteriors estimated by v6 are very flat (all parameters have flat posteriors) resulting in very high variances while the same model estimated under v5 seems to return more “normal” results. Has the sampling algorithm changed?
You are most likely falling victim of the bug with the penalty in the objective function that appears in versio 3.062. I hope to provide a new version soon. The SVN version of version 3 is corrected. Use the download instructions for version 4, replacing dynare_v4 by dynare_v3