fpp3: Chapter 13.4 Forecast combinations

[holleland]

Hi, I came across an issue in Ch 13.4 of the fpp3 book. I guess something has changed in the packages being used, because the combination forecast is not an average of the other forecasts. Looks like it would need a level shift. From fiddling with the code, it may seem that when the combination forecast is being made, the STLF model is contributing to the aggregated forecast on the log-scale. If I manually calculate the mean of the .mean forecasts, I get the correct values for the combination forecast.

I think the only change needed is to replace
mutate(combination = (ets + stlf + arima)/3)
by
mutate(combination = mean(ets, stlf, arima)).

At least, I then get:

which I think is what this figure used to look like, the last time I gave my course.

Thanks for a great textbook that I use in my forecasting course.

Best regards,
Sondre Hølleland

[robjhyndman]

Thanks. This is a bug in fabletools. Will fix in the next release. See tidyverts/fabletools#443