kalman filter

[Lia]

I am trying to decompose the gdp into a cyclical and a trend component. I am using the following specification:
@signal dlgdp = sv1+ sv2

@state sv1 = sv1(-1) + [var = exp(c(1))]
@state sv2 = c(3)*sv2(-1) + [var = exp(c(2))]

Where dlgdp is the growth rate of the GDP.
The last week I was able to do it, nevertheless I re try to do it today and it give a warning message that say "singular covariance", but I am using the same data and the same data.

What can I do?

[startz]

I am trying to decompose the gdp into a cyclical and a trend component. I am using the following specification:
@signal dlgdp = sv1+ sv2

@state sv1 = sv1(-1) + [var = exp(c(1))]
@state sv2 = c(3)*sv2(-1) + [var = exp(c(2))]

Where dlgdp is the growth rate of the GDP.
The last week I was able to do it, nevertheless I re try to do it today and it give a warning message that say "singular covariance", but I am using the same data and the same data.

What can I do?

Change the starting values in the C vector before you estimate.

[Lia]

Thank you for your reply!
How Can I change the coefficient in the c vector?
It is in View - specification - coefficient value?

Because even thought I can see the values I can not modified them...

Thank you!

[startz]

Open the C object in the workfile and edit the entries as you would any other object

[Lia]

for "C object" , you refer to View - specification - coefficient value? in the state space window right? I have modified the coefficient in the coefficient values matrix mentioned, and it still give the same error.

I am sorry to do not get it so quickly. it Is my first week with the state space section of eviews...

[startz]

for "C object" , you refer to View - specification - coefficient value? in the state space window right? I have modified the coefficient in the coefficient values matrix mentioned, and it still give the same error.

I am sorry to do not get it so quickly. it Is my first week with the state space section of eviews...

No, I meant in the workfile window.

[Lia]

Finally I get it!!!! :D

I fillout the C Vector in the workfile with zeros.. is that correct?

and it give me the same result

WARNING: Singular covariance - coefficients are not unique

[startz]

Finally I get it!!!! :D

I fillout the C Vector in the workfile with zeros.. is that correct?

and it give me the same result

WARNING: Singular covariance - coefficients are not unique

Yes, that was correct. You may want to check the results to see if the coefficients go to plus or minus infinity. You might also want to post the workfile to see if anyone knows what the problem is.

If you will forgive what may seem an odd question, are you sure you intend to model dlgdp rather than log(gdp)? I ask because GDP has a large unit root component. I'm not sure the growth rate does.

[EViews Gareth]

It may be that zeros are bad starting values.

Try other numbers, either at random, or ideally with some idea of what they should be.