I have a question regarding the estimation approach of Guerrieri and Iacoviello (2017) for piecewise linear models (i.e. Occbin plus estimation). My understanding is that unlike the Kalman (or particle) filter, this approach cannot deal with having fewer shocks than observables. I was wondering whether it could in principle deal with missing observations for an observable? Or does that not work either?
That should also not work, because you are essentially inverting the model to get the shocks. With missing data, the inversion cannot be done.
Even with the Kalman filter, you need at least as many shocks as observed variables.
Thanks Michel and Johannes, I apologize there was a typo in my post, it should be “My understanding is that unlike the Kalman (or particle) filter, this approach Guerrieri and Iacoviello (2017) cannot deal with having MORE shocks than observables.”
But maybe you guessed that already.
Regarding the inversion with missing observations, I suppose one could do the inversion if one were willing to switch off a specific shock during that period when he observation is missing?
Yes, that should work.