Say we have a bunch of entities that develop over a time-series in n-steps in a fashion that we hope is similar but just attenuated by some other entity-specific features. The data structure is the same for each entity and said features.
Would it be appropriate to:
- Train a time-series approach (perhaps a VARIMAX? I am not super familiar here) on a collection of entities and how their time-series played out
- Apply that model to new, not-yet-seen, entities and predict how they will play out
I'm more used to typical regression and would think just including some lag columns might be more appropriate here, but I'm not sure if time-series approaches would offer something more.