FPP3 hiearchical forecasting guidance?

Previously read FPP2, about to start FPP3 as a segue to using #fable, #tsibble , … #tidyverts . Noticed fpp3 doesn't have chapter on hierarchical forecasting (though FPP2 did). Curious if planning to add this in the future? Or if there are any tutorials available for applying hiearchical forecasting techniques with fable?

Referred here by Forecasting: Principles and Practice, by Rob J Hyndman and George Athanasopoulos

There's not much on rseek.org using fabletools hierarchical as the search term. the FFP3 authors have a paper out on the principles that came out recently. Glancing at the fabletools documentation, it seems like there should be the functions to implement it, but that's only an impression (wild guess?).

@technocrat thanks for the link, looked into the code for that paper and looks like it's using more forecast and gts package approaches still (code example). Yes, it seems like the pieces are there, can also see presentation materials here.

(Am wondering if they are finishing up some of the distributions w/ reconciliation stuff and then will put out more documentation walking through hierarchical and grouped contexts...)

1 Like

We will add a chapter on hierarchical forecasting to fpp3, but it is not yet done. One of the reasons we have delayed is that we may change the syntax for forecast reconciliation to make it more efficient.

However, there are examples in the help files for reconcile() and aggregate_key() which shows how it can be done using current package versions.


Now that we've heard from The Man Himself it would be a kindness for future pilgrims to mark his answer as the solution. Thanks.

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.