EViews.com

Dear Forum Participants,

I'm Rika. I'd like to use Kalman Filter model for generating the volatility series of eight financial variables. Each variables will be estimated separately. One of the variables that will be estimated is the banking non performing loan (NPL).What I understand about Kalman Filter for generating the volatility series is: firstly, I estimate the filtered series of the NPL (say, the NPLbar -- using Kalman Filter model**), then I count the squared difference of the filtered series (NPLbar) from the variance of the real data (NPL) ==> (NPL-NPLbar)^2. Am I right in formulating this steps for generating the NPL volatility series?

**For estimating the filtered series (NPLbar), I'm using the following model:
@signal NPL=sv1+[var=exp(c(1))]
@state sv1=sv1(-1)+[var=exp(c(2))]

Has this been a good model? Is there any guideline for me to formulate the Kalman Filter model, since I just want to get the filtered series of the NPL?

My next question...
for generating the filtered series, which one should I follow:
1) Proc/Make State Series/..., or
2) Proc/Make Signal Series/...


Thanks in advance
Rika :)

Kalman Filter is a new topic for me.. Could anyone give any suggestions for me to solve my problem about Kalman Filter?

The specification you formulated is called "local level model". Depending on the properties of your variables, you may also want to specify "local level trend model". If you do a quick search in the forum, you'll find useful examples and discussions on similar issues. Regarding to your last question, you should go for Proc/Make State Series/...

If you are truly willing to deal with such applications of Kalman filter, I suggest you to read (at least first few chapters of) "An Introduction to State Space Time Series Analysis" by Jacques J.F. Commandeur and Siem Jan Koopman (2007).