deterministic trends

Topics: Kalman Filtering, Models
Sep 17, 2015 at 12:45 AM
How do I add deterministic trend to measurement variable in a stationary model?
!transition_variables x
!measurement_variables y
!transition_equations
x = 0;
!measurement_equations
y = x;
!dtrends
y += 1;
get(model('test.model'),'dtrends') produces
    y: 1
    x: 0
but I want
    y: 1i
    x: 0
Thanks,
Iskander
Coordinator
Sep 17, 2015 at 5:42 AM
The dtrends block affects the measurement block, not the state block. That is, specifying a deterministic trend there will detrend the data before it hits your measurement block but will leave your model steady state otherwise unchanged.

If you look at the tutorial Jaromir's simple SPBC model you will see the block used to de-mean observables. This is because there is just a single parameter in each of the equations. If you want a linear time trend then you need to use the !ttrend keyword. But again, this isn't going to affect your steady state.

If you want growing variables in your steady state, then you need to setup your mode in such a way that this is implicit in the equations. If IRIS is having trouble figuring out the steady state and growth rates you want (possibly there is more than one solution for a given structure) then you can use the fixGrowth= option when calling sstate. Note, however, that things are in decimal notation, so 1i would actually imply 100% growth for that variable period-on-period in steady state. If you want 1% period-on-period growth in that variable, it should be 1+0.01i.
Marked as answer by jaromirbenes on 9/17/2015 at 12:03 AM
Coordinator
Sep 17, 2015 at 7:03 AM
Hi Iskander

The syntax
y += 1
means that a constant of 1 is added to a measurement variable before it gets observable or confronted with observables (e.g. in the Kalman filter). To add a time trend, you need to use e.g.
y += a*!ttrend
In general, you can also use your own series in the !dtrends blocks (exogenous variables), as in
!exogenous variables
xxx

!dtrends
x += a*xxx;
and then simply supply a time series for xxx in the input database.

Hope this helps.
Best,
Jaromir
Marked as answer by jaromirbenes on 9/17/2015 at 12:03 AM