forecast package: accuracy and Box-Cox

I have two questions concerning the forecast package:

  1. accuracy-function - what does it do? In my example (see below) the test set is of length 24 month and the forecast-intervall 6 month. So what does the function „accuracy“ do? Does it run a rolling window (of size 6) with step 1? Or does it a stepwise forecast with step 6 (here 4 forecasts)? Or something else?

Assume we have data („Data“) up to 2020,12 and want to evaluate the forecast performance with a training set (up to 20018,12) and a test set (up to 2020,12) like this:
train_Data <- (window(Data, end=c(2018,12))
fit.model <- ets(train_Data)
test.model <- fit.model %>% forecast(h=6) %>%
accuracy(Data)
test.model[,c(„RMSE“,“MAE“)]

  1. accuracy and Box-Cox - are the RMSE calculated with the back-transformed data?

If we fit a model with the Box-Cox transformation (forecast function: lambda=“auto“) and evaluate the forecast with the accuracy function: Are the RMSE, MAE, etc. calculated with the Box-Cox transformed data or with the back-transformed data? So, are the RMSE, etc. comparable with those from models without Box-Cox transformation?

Thank you very much!

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.