Filtered variables with particle filter

Estimation
1

Hi!

Is it possible to construct a filtered series of the shocks after running the particle filter (order 3)?

I understand that the smoother is not available, but what is the closest object that can be computed with the filter?

Thanks!

2

With the default settings that is not yet feasible. Which filter type are you using?

3

Do you have suggestions on how to store the filtered series? As I understood, they are part of the intermediate calculations. These are the options for the filter, but I can change them if it helps:

estimation(order = 3,

        nonlinear_filter_initialization=3,

        datafile = 'Output/Data', diffuse_filter, mh_nblocks=1, mh_replic=100,

        mh_init_scale=4, mode_compute=5, mode_check,

        posterior_graph, mh_jscale=1, plot_priors=1,

        mode_check_neighbourhood_size=2, particle_filter_options = ('pruning', false))

                                    pi_q g_q;
4

I saw that you are using a smoother in the JME paper (GitHub - JohannesPfeifer/Particle_Filtering: Matlab Particle Filtering and Smoothing Example Code · GitHub). Does this provide smoothed series for the shocks and variables?

5

Yes, but only for a simple AR process. I am currently working the Dynare implementation. For the filtered series that is straightforward. I hope to be able to provide something later this week.

1 Like
6

Thanks, this is really helpful.

In my model, the exogenous shocks follow simple AR(1) processes. The endogenous variables do not. Is this case applicable to the GitHub smoother?

7

Yes, it should be.

1 Like