uncertain futures about greykite

[al-bert]

Great question! (and devastating penalty shootout for France a few days ago..)

In this case, you would make multiple forecasts with greykite (and any other library) and then apply your own method to generate the consensus forecasts.

It might be possible to stretch the usage of ReconcileAdditiveForecasts to generate such consensus, by creating a "tree" where each parent has a single child, but we haven't tried it ourselves. The consensus forecast would then be a linear combination of the input forecasts. See #25 for a brief description of that method if interested. Before going too deep into this, I'd recommend starting with a simpler method like median, as you mentioned, and making sure your input forecast are good.