I noticed that in 3_064, each call to the forcst2 function in metropolis.m asks for only one forecast. ie:
yf1 = forcst2(yyyy,horizon,dr_,1);
In previous versions, the last argument passed to forcst2 was “B” (which at the time was set to 100, currently 500) instead of 1. Is there a reason that this version uses only 1 draw of the future shocks to determine the forecast uncertainty from the parameters and the future shocks?
Yes, this call takes place in a loop over B draws among metropolis iterations.
At first, I thought it was necessary to sample more for parameter uncertainty and intrinsic shocks, then I revised this opinion. I don’t think anymore it is necessary to do B draws to describe parameter uncertainty and B*B draws for parameter uncertainty and intrinsic shocks
Of course, we should make B a parameter that users can tune. But as you have found it, you can now tune it by hand