the manual (section 4.13.1) says that the simulated series requested by the “simul_replic = INTEGER” option are not used for the calculation of the empirical moments. The moments of my model, however, change significantly when I change the number of replications.
Is it possible that the reported empirical moments are actually averages over the number of replications?
No, but the problem is that empirical moments are computed from the last one of the replications. If you increase simul_replic, the last run will be different as it uses different random numbers.
I have a related question. Do contemporaneous correlations work the same? Are the elements of the matrix averages of the correlation coefficients over the number of replications set in simul_replic or are they just the coefficients of the last replication?