Page 1 of 1

Unobserved components models with lagged dependent variable

Posted: Sun Mar 22, 2015 8:09 pm
by Yohan
Hi everybody!

I want to estimate a unobserved components model with explanatory variables.
Thereby the explanatory variables include also a lagged dependent variable y(t-1) (the remaining terms are other lagged variables of order (t-1)).

By means of the aforementioned model i want to calculate one-step-ahead out-of-sample forecasts with recursive parameter estimation.

I know i can manage a recursive forecasting scheme with time-varying parameters including in the state vector.
The usual approach is to specify a random walk specification for the parameters.

But is this viable for the parameter of the lagged dependent variable?
Literature suggests, that time-varying parameters of lagged dependent variables need some degree of continuity.

Are there specifications which are more suitable?
I thank you very much for a discussion.

Many greetings from Germany, Yohan.

Re: Unobserved components models with lagged dependent varia

Posted: Mon Mar 23, 2015 1:39 am
by trubador
Yes, random walk specification is very flexible, which may not be appropriate for certain cases. You can also try AR(1) or smoothed spline specifications.