Conditional_forecast: why is the unconditional forecast not the same as when using the forecast option in estimation?

Good evening,
I have a question regarding the conditional_forecast command in dynare 4.5.7. Why is the unconditional forecast stored in forecasts.uncond.Mean not identical to the one I get when using the “forecast=integer” option in the estimation command? I did not run an MCMC as part of the estimation, so there is only the posterior mode obtained with mode_compute, and as you can from the code fragment I pasted below, I did choose parameter_set = posterior_mode. So is the forecast reported in forecasts.uncond.Mean not a forecast computed at the posterior mode?

Furthermore, what is the replic option for (it is not clear from the manual). It appears that the larger value I set for replic, the closer the forecast in forecasts.uncond.Mean moves towards the forecast obtained with the “forecast=integer” option in the estimation command. Is it that the conditional_forecast performs random draws of the estimated parameters, the total number set in replic, and performs a forecast for each of them? And if so where is it drawing them from (since I have not conducted an MCMC)? A multivariate normal using the mode and estimated variance-covariance matrix of the parameters?

Many thanks for your help,

var dEXCRN;
periods 1:12;
values 0;

conditional_forecast(parameter_set = posterior_mode, controlled_varexo = (EE_S_usd), replic = 20000);

Could you maybe provide me with the codes to investigate?

Hi Johannes, I am so sorry,
it’s a pretty long code so I was hoping to replicate the issue in a mickey mouse model, and then other things intervened. I am hoping to get back to this in two weeks time.