singular covariance in Kalman filtering

For questions regarding programming in the EViews programming language.

Moderators: EViews Gareth, EViews Jason, EViews Moderator, EViews Matt

remuct
Posts: 27
Joined: Fri May 12, 2017 6:47 am

singular covariance in Kalman filtering

Postby remuct » Wed Nov 30, 2022 3:04 pm

Hi,
I have a concern about a "singular covariance" message which came with the output of a Kalman filter estimation. I am wondering which covariance is concerned here, the covariance of the measurement or of the state equation disturbances or the covariance of the forecasted or updated state?
Maybe this can help: the variable x(t) in my signal eqn y(t)=a(t)x(t)+ u(t) is a weighted average of actual and past values of a variable z(t) such that x(t)= (b1 z(t)+ b2 z(t-1))/(b1+b2). When I specify x(t) this way, I get the message singular covariance. When I specify it equivalently by dividing all parameters by b1, i.e. as x(t) = (z(t) + c z(t-1))/(1+c), I do not have the message anymore. Could it be that the first specification leads to some overidentification problem in the signal eqn, which would explain the singular covariance message of Eviews? And again, what is the covariance concerned?
Thank you for your helpful comments.

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: singular covariance in Kalman filtering

Postby EViews Glenn » Wed Nov 30, 2022 3:54 pm

It is the covariance of the coefficients.

It is a little difficult to diagnose without looking at the full specification and data and thinking a lot more than I am doing at present, though I think you are right about overidentification. I will note that in the second (non-singular) parameterization, you are only estimating a single share coefficient instead of the two levels. Which leads me to believe that there is an overidentification problem where the levels are set somewhere else in the specification so all that you can identify is b2/b1.


Return to “Programming”

Who is online

Users browsing this forum: No registered users and 25 guests