In our estimations, the “oo_.SmoothedVariables” output does not behave as described in the manual.
The manual says about oo_.SmoothedVariables:
[quote]Variable set by the estimation command (if used with the smoother option), or by the calib_smoother command.
After an estimation without Metropolis, or if computed by calib_smoother, fields are of the form:
After an estimation with Metropolis, fields are of the form:
We run “estimation” without the “smoother” option (and with Metropolis draws). Still, our oo_ output contains both oo_.SmoothedVariables.VARIABLE_NAME and oo_.SmoothedVariables.MOMENT_NAME.VARIABLE_NAME fields. This deviates in two respects from the manual: (1) there is output even without the smoother option, (2) both types of output co-exists (whereas the manual suggests that they should be mutually exclusive).
Is this a bug or can this output be meaningfully used?
Now my confusion is complete:
When I run the estimation with the smoother option, I still get both types of fields, but now the content of oo_.SmoothedVariables.MYVAR and oo_.SmoothedVariables.Mean.MYVAR is exchanged compared to the case without the smoother option.
Further examination has revealed that the output reversion problem from post #2 appears only with one particular variable, which is the same variable that causes the problem in the thread Latent variables with data. Therefore the problem is most probably specific for our model context and this particular variable. I’ll sent up the model code.
Are you using the unstable version? And is your model maybe stochastically singular?
No, I’m using version 4.4.3 and I’ve not encountered singularity problems with my model.
When I prepared the code for sending it up and double-checked whether I was able to reproduce the results, I found that I was not. This gave me the clue to what had gone wrong. Apparently, I was partly looking at results of older estimation runs that were still included in the oo_ variable had not be overwritten or deleted.
And the reason for this, in turn, was that I was routinely using the “noclearall” option in invoking Dynare (following the suggestion in the thread Strange warning message for avoiding spurious warnings). Apparently this caused old fields of the oo_ variable to remain.
Without the “noclearall” option, the oo_ content looks much more reasonable:
[ul]]If I run “estimation” without the “smoother” option, I get oo_.SmoothedVariables.VARIABLE_NAME output, but no oo_.SmoothedVariables.MOMENT_NAME.VARIABLE_NAME output./:m]]If I run “estimation” with the “smoother” option, I get oo_.SmoothedVariables.MOMENT_NAME.VARIABLE_NAME output, but no oo_.SmoothedVariables.VARIABLE_NAME output./:m][/ul]
So the only question that remains is: What is the interpretation of the oo_.SmoothedVariables.VARIABLE_NAME output if I run “estimation” without the “smoother” option, but with Metropolis draws?
- Please upgrade to the unstable version due to the various bugfixes available (dynare.org/DynareWiki/KnownBugs). It will be released as Dynare 4.5 in the next month or so.
- From the bug list
[quote]ESTIMATION: Results displayed after Bayesian estimation when not specifying the smoother-option are based on the parameters at the mode from mode-finding instead of the mean parameters from the posterior draws. This affects the smoother results displayed, but also calls to subsequent command relying on the parameters stored in M_.params like stoch_simul (a workaround can be found here)