Where are the posterior theoretical moments (mean, std) saved (apart from the log file)? I conjecture that the mean should be taken from the oo_.steady_state, although it would be nice to see it in the oo_.PosteriorTheoreticalMoments as well. But I cannot find the std: I presume it should be in the oo_.PosteriorTheoreticalMoments.dsge.covariance mean' orvariance’ structure but both deliver NaNs for all variables although the log file gives reasonable numerical values.
Additionally, (the same mod-file but should be under a different subject) I’m asking the estimation command for the k-step ahead forecasts but no output is provided in the oo_ structure:
estimation(order=1,datafile=ctwBaselineData, mh_replic=300000, mh_nblocks=1, mh_drop=.3, mh_jscale=0.2,
mode_compute=6, filter_step_ahead = [1 4 8 12], forecast=20,smoother, filtered_vars, diffuse_filter,
nodisplay,
% do this only for the final model b/c takes a lot of time
moments_varendo,
conditional_variance_decomposition=[1 4 8 20],
%
graph_format=(eps,pdf))
baseline70log.txt (432 KB) baseline70.mod (57.7 KB)
Still with ver.4.3.3.:
As far as I understood, this happens when some of the variables are nonstationary, i.e., would result in the `NaN’ value. Then, the log-file reports NaN values only for the nonstationary variables but the oo_ structure yields all NaNs regardless of stationarity.
Repeating the exercise without the nonstationary variables, the oo_ structure gives numerical values. However, the values in the oo_.PosteriorTheoreticalMoments.dsge.covariance.mean' structure slightly differ from the ones reported in the log-file under the titleTheoretical moments’.
Theoretical moments are displayed only in the log file for stoch_simul. Those moments are based on the posterior mean of the parameters when called after estimation.
But the posterior moments stored in oo_ are the mean of the posterior moments (given parameter draws from the posterior distribution). This is different than the moments at the mean of the parameters.