Option g in DLM

[hardmann]

Hi Tom,

I have been trying to imitate @disaggregate to solve my problemhttps://estima.com/forum/viewtopic.php?f=13&t=3433. I found in this procedure, it use g to handle non-stationary states.Since I have mixed stationary and non-stationary states and I tried option PRESAMPLE=ERGODIC and CONDITION=3which doesn't work well, I wonder whether g can handle this mixed situation.If it can,how does g work? OR option PRESAMPLE and CONDITION is correct choice but I didn't do it the right way？

With much appreciation,
H

[TomDoan]

PRESAMPLE=ERGODIC should work if you get everything else correct, though you need to be careful that the AR cycle process doesn't pick up a unit root itself. (That's a common side effect of having too flexible a UCM). However, the %DLMGfromA(A) function will give you a (one of many) G matrices. In the case of your model, the tau's and d will have tau(t)-tau(t-1)-d(t) and tau(t-1)-tau(t-2)-d(t) as stationary linear combinations (and, of course, the cycle states are supposedy stationary).