Dummy variable in state equation

[fernando.88]

I am trying to add a dummy variable to a state equation on a state space model.

The estimation is successful (in terms of convergence), but the estimated coefficient for the dummy variable is not what I expected it to be. I tried looking at the predicted and filtered states and the only way I can understand the estimated coefficient depends on the following being true: on the state equation @state x = x(-1) + dummy, the predicted state in time t depends on the dummy variable on time t-1. This interpretation seems to be consistent with the user guide description of the state equation, where the left hand side is indexed on time t+1 and the right hand side is indexed on time t. Is this interpretation correct? My fisrt understanding of the notation used to specify the SS object was that the predicted state in time t depended on the dummy variable on time t.

I redefined my dummy variable by shifting it back 1 period and now the estimated coefficient is consistent with what I expected.

I am using Eviews 8.1.

Thanks.

[trubador]

Yes, your reading and the proposed solution is correct.

[fernando.88]

Thank you, trubador.

I am not sure if this is a common misunderstanding of the SS notation, maybe it would be a good idea to highlight it on the user guide. I only noticed the behavior because the exogenous variable was a dummy, which made it simpler to interpret.

[trubador]

Actually, there is a small section in the manual that highlights the issue: http://www.eviews.com/help/EViews%209%2 ... l#ww159573