As much as I love the UC framework, there are some glaring concerns...
[ ] unclear support for multivariate models
[ ] smoothing is broken since base operators aren't generally defined
[ ] UnobservedComponent is currently defined specifically for given models
I really want to emphasize that last one. If this can be generalized in any way, then we can easily address the second point and make it even more appealing to the laymen.
Lastly, multivariate support is a tough sell. We would need to divide the UnobservedComponent into idiosyncratic series, since each one has it's own trend process. Furthermore, there could be common components as is the case with (Creal-Koopman-Zivot, 2010).
As much as I love the UC framework, there are some glaring concerns...
UnobservedComponent
is currently defined specifically for given modelsI really want to emphasize that last one. If this can be generalized in any way, then we can easily address the second point and make it even more appealing to the laymen.
Lastly, multivariate support is a tough sell. We would need to divide the
UnobservedComponent
into idiosyncratic series, since each one has it's own trend process. Furthermore, there could be common components as is the case with (Creal-Koopman-Zivot, 2010).