Uncond Forecast and forecast opt yield different results

Dear all,
I am quite puzzled since I found that the function imcforecast.m in its unconstrained version yields confidence intervals that are different than the ones produced (at the same significance level) with the forecast option in the estimation command. To reproduce what I got, consider as an example a code that I attached (it was provided in a previous post Some questions, few answers) which I changed accordingly (I ran it in dynare 4.0.4). After running the example just type:
load conditional_forecasts.mat
forecasts.uncond.ci.y
oo_.PointForecast.HPDsup.y
oo_.MeanForecast.HPDsup.y
and it will become apparent that the first two vectors differ, contrary to what I would expect! In general, what i found is that the forecast option yields wider confidence intervals. Why is it the case that the forecast uncertainty is different in these two alternative ways? Or I am messing thins out? can you explain?
To make a more careful investigation, can you tell me where vectors that represent green lines in forecast graphs are stored. Then I can loop at different values, ie, 0.1…0.9 and compare with those that result from imcforecast.m
Many thanks in advance for your attention. Best, jorge
New Folder.rar (6.37 KB)

Dear all,
An update, I found out that all the issue of the forecast option in estimation appears in line 961 in “dynare_estimation1.m” which calls “prior_posterior_statistics.m” that make the summary of stats among all the MH draws (or line 1461 if you load your previous mh draws). After some manupulations you arrive at lines 200-201
stock_forcst_mean(:,:,irun(6)) = yf'; stock_forcst_point(:,:,irun(7)) = yf1';
Then takes subdraws and finally calls “pm3.m” which makes the forecast and conf intervals out of the draws. Or manually you can get the green lines (percentiles) manually typing:
plot(1:13,oo_.MeanForecast.Distribution.p,’-g’,‘linewidth’,0.5)
plot(1:13,oo_.PointForecast.Distribution.p,’-g’,‘linewidth’,0.5)
where the number 13=options_.forecast+1

But still I remain dubious with the comparison of with (unconditional) mcforecast3. Why latter’s confidence intervals have lower coverage than those of the forecast option in estimation? (If both methods use the same Kalman transition eq to make the forecasts, then it follows that variances should be different)
A related question is which command in dynare takes advantage of forcst_unc.m?
Any clarification is very much appreciated. Many thanks in advance, Jorge