Page 1 of 1
Forecasting with SSpace
Posted: Wed Apr 14, 2010 12:22 pm
by fmgoto
Hi, I have defined a given SSpace and I am trying to forecast 3-periods ahead with observed explanatory variables. However, it seems that Eviews does not forecast if the SSpace's signal contains lagged dependent variable.
KALMAN.APPEND @SIGNAL LOG(Y) = SV1 + SV2 * LOG(Y(-1))
KALMAN.APPEND @STATE SV1 = SV1(-1) + [VAR=EXP(C(0))]
KALMAN.APPEND @STATE SV2 = SV2(-1)
Is this so? Any idea on how to accomplish the forecast?
Thanks.
Re: Forecasting with SSpace
Posted: Wed Apr 14, 2010 1:37 pm
by EViews Glenn
I'm not sure why this doesn't work for you. What is the error message that you get...
Re: Forecasting with SSpace
Posted: Thu Apr 15, 2010 5:18 am
by fmgoto
Glenn, I got no error. I just didnt get the result I expected.
Given that I do have "x" until 2010m03, I was expecting to get "y_hat" until 2010m03 and not 2009m12. Isnt that so? Why?
Thank You, Fabio
Code: Select all
smpl 2000m01 2009m12
sspace KALMAN15
KALMAN15.append @signal log(y) = sv10 + sv_ar1 *log(y(-1)) + sv11*dlog(x)
KALMAN15.append @state sv10 = sv10(-1) + [var = exp(c(10))]
KALMAN15.append @state sv_ar1 = sv_ar1(-1)
KALMAN15.append @state sv11 = sv11(-1)
KALMAN15.ml
smpl 2000m01 2010m12
KALMAN15.forecast(i=o, m=n) @signal y_hat @state sv10_in sv_ar_in sv11_in
Re: Forecasting with SSpace
Posted: Thu Apr 15, 2010 9:53 am
by EViews Glenn
It's hard to say without seeing your data, but I will note that the following paragraph from the manual may be relevant here
Signal equations may not contain current values or leads of signal variables. You
should be aware that any lagged signals are treated as predetermined for purposes of
multi-step ahead forecasting (for discussion and alternative specifications, see Harvey
1989, p. 367-368).
Re: Forecasting with SSpace
Posted: Tue Apr 20, 2010 11:11 am
by fmgoto
Thank You!