Hi All
I wonder if anyone has experimented with the fast_kalman_filter option for Dynare estimation. When I tried with a smaller model, it does not seem to make much of a difference in terms of computational speed. Perhaps with bigger models it will be different? Keen to know if somebody has used this option.


I tried it and for some models it makes a big difference, while for others it hardly makes a difference at all. This is actually consistent with what Ed Herbst describes in his paper. But the gain should be bigger for big models where the number of states is typically much bigger than the number of observables.

1 Like