Incorrect pruning at order 2 with particle filtering
An anonymous user noticed that our implementation of pruning at order 2 with particle filtering is probably incorrect. The problem is located in MEX local_state_space_iteration_2
.
More precisely, when computing next period pruned variables, instead of adding the term +ghxu·ŷ₁⊗ε
(as indicated in the comment), the code actually adds +ghxu·ŷ₂⊗ε
. See https://git.dynare.org/Dynare/dynare/-/blob/master/mex/sources/local_state_space_iterations/local_state_space_iteration_2.cc#L118
The paper by Kim et al. (2008), and also our MATLAB/Octave implementation of pruned simulations (in matlab/simult_.m
), both use ŷ₁
at that place.
I’d be grateful if someone else could confirm this diagnostic. Cc @stepan-a @JohannesPfeifer @normann