EViews.com

Hello,

I am trying to find if kalman filter can be used for calculating parameters for combined forecasts.

For example if we have f1,f2 forecasts based on Granger and Ramanathan (1984)
we can calculate a forecast combination of the form of a regression. fc= a0 + a1*f1 + a2*f2+ u. So we are trying to find the parameters of a regression like Y=A0+A1*f1+A2*f2.

Can we calculate the coefficients A0,A1,A2 as adapting weights of the kalman type?

And if we can what type of data do we need ?

I appreciate any help on the Kalman filters because I am quite lost here!!

Can we calculate the coefficients A0,A1,A2 as adapting weights of the kalman type?

Yes, you can. This is called "time varying parameter" estimation. You can define the coefficients as state variables or as Stochastic Regressors in EViews' terms. EViews offers you four choices:
1) Constant mean (plus noise): A1t = b0 + et
2) AR(1) coefficients: A1t = b0 + b1*A1t-1 + et
3) Random walk coefficents: A1t = A1t-1 + et
4) Random walk (with drift) coefficients: A1t = b0 + A1t-1 + et

Since it is an empirical process to determine the right specification for your data, you should try them all. However, you can choose a specific one if you have prior belief or a theoretical reason to do so.

And if we can what type of data do we need ?

No additional data is needed as long as you have fc, f1 and f2.

I appreciate any help on the Kalman filters because I am quite lost here!!

Kalman filter analysis is not an easy task, so you should study its theory before going any further. Moreover, you should also refer to EViews' manual to understand how to carry out Kalman filter analysis in EViews. If you encounter any problems in the estimation process, please search the Forum for possible solutions and advices.

Dear friend,

Thank you so much for the response.

I looked up the kalman filter theory from eviews help guide.
But I am having difficulties associating my data with the complexicity of the kalman filtering matrixes.

I am planning to use eur/usd ratio as time series to perform several forecasts.

I am not sure that I make any sense, but that's the main problem on my mind!