Possible to acquire prediciction intervals for reconciled prophet forecasts?

[henningsway]

Thank you for bringing prophet to fable!

I plan on using prophet in a hierarchical forecasting context, where I require reconciled forecasts. I'm not too familiar with the concepts and methods yet, but I was hoping, that also the reconciled prophet-forecasts would have a distribution (as it seems to be the case for ETS() for example), potentially generated via bootstrapping forecast errors.

Is it possible to get a prediction interval for reconciled prophet forecasts? :)

Please finde a reprex below for the approach I have taken so far.

library(tsibble)
library(fable)
library(fable.prophet)
library(dplyr)


tourism_states <- tourism %>%
  aggregate_key(State / Region, Trips = sum(Trips))

mdl <- tourism_states %>% 
  model(prophet = prophet(Trips)) %>% 
  reconcile(prophet_mint = min_trace(prophet, method = "mint_shrink")) %>% 
  select(-prophet)

fc_rec <- mdl %>% 
  forecast(h = "1 year")

fc_rec %>% 
  filter(is_aggregated(State)) %>% 
  autoplot(tourism_states)

fc_rec_b <- mdl %>% 
  forecast(h = "1 year", bootstrap = TRUE)
#> Error: Problem with `mutate()` input `prophet_mint`.
#> x unbenutztes Argument (bootstrap = TRUE)
#> i Input `prophet_mint` is `(function (object, ...) ...`.

fc_rec_b %>% 
  filter(is_aggregated(State)) %>% 
  autoplot(tourism_states)
#> Error in eval(lhs, parent, parent): Objekt 'fc_rec_b' nicht gefunden

^{Created on 2020-12-10 by the reprex package (v0.3.0)}

[mitchelloharawild]

Probabilistic reconciliation is currently only supported for normally distributed forecasts, however is possible to also reconcile the sample paths to give coherent intervals. This will need to be added to {fabletools}.

[henningsway]

Thanks, it's good to know, that this functionality ist not yet available (and that it's not me using the package in the wrong way).

Do you see a way of estimating a normal distribution from the prophet forecast and maybe use this as an input for the reconciliation or some other kind of workaround?