topic Re: implementing kalman filter when Ht (the measurement matrix) is time varying in SAS/IML Software and Matrix Computations

implementing kalman filter when Ht (the measurement matrix) is time varying

tarikb02 — Thu, 02 Feb 2012 12:16:12 GMT

Hello,

I would like to implement kalman filter when the measurement matrix is time varying. the model is as follows:

Yt = Ht * Zt + Et Ht = [ 1 Mt]

Zt = Zt-1 + Vt.

can anyone tell how to do this with KALCVF in iml. My problem is that my Ht is time varying while in KALCVF I need Ht is time invariant.

Many thanks in advance.

tarik

Re: implementing kalman filter when Ht (the measurement matrix) is time varying

Rick_SAS — Fri, 06 Apr 2012 18:19:38 GMT

A colleague sent me the following information regarding this question:

Actually the documentation does mention that a time-varying
measurement matrix, H_t, is permitted. All you need to do is provide H_t
that is a ((T+lead)*N_y) times N_z matrix where, T is the number of time
points, lead is the forecast lead, N_y is the dimension of the observation
vector, and N_z is the dimension of the state vector. For
example, in the user’s notation, if the observations are univariate, the
rows of H_t matrix will be [1 M_t], t=1 …(T +lead).

By the way, as a side note, mention to the user the new state
space modeling procedure, SSM, in 9.3 (its doc is at http://support.sas.com/documentation/cdl/en/etsug/63939/HTML/default/viewer.htm#etsug_ssm_sect006.htm
).