by jpfeifer » Sun Apr 21, 2013 8:00 am
Given the state space representation of the model, Dynare uses the Kalman filter/smoother to compute forecasts. The "mean forecast" distribution is generated by taking draws for the parameters distribution generated in the MCMC and computing the forecasts for those draws. Given those forecasts, you can now compute statistics like the median or the 90% HPDI/percentiles. Does this answer your question?