Page 1 of 1

Kalman Filter: Asymmetric covariance matrix

Posted: Mon Jul 29, 2019 10:46 am
by pcruzd
Hi I am estimating a State Space model that has two state variables x and y, with errors res_x and res_y. res_x variance is not affected by res_y, but the variance of res_y is affected by res_x. However, i am not sure how to estimate such a structure in Eviews. What I am looking is to specify the model such as:

Code: Select all

@ename res_x @ename res_y @evar var(res_x) = exp(c(1)) @evar var(res_y) = exp(c(2)) @evar cov(res_x, res_y) = 0 @evar cov(res_y, res_x) = c(3)
I know the latter is wrong as the cov(x,y) command is symmetric, so I need to know if there is any way to compute such assymetric covariance matrix.

Re: Kalman Filter: Asymmetric covariance matrix

Posted: Mon Jul 29, 2019 11:01 am
by startz
By definition, a covariance is symmetric. Nothing to do with EViews.

Sounds like you may want to build a model of the errors explicitly into the state space model.

Re: Kalman Filter: Asymmetric covariance matrix

Posted: Mon Jul 29, 2019 12:02 pm
by pcruzd
you are right, already modelled the errors explicitly. Thanks!