Compute scaled errors for accuracy by series, then summarise.

This doesn't look right:

library(fpp3)

fc <- aus_arrivals %>%
  filter(year(Quarter) <= 2009) %>% 
  model(
    ets = ETS(Arrivals),
    arima = ARIMA(Arrivals)
  ) %>%
  forecast(h=11)
fc %>% 
  accuracy(aus_arrivals) %>%
  select(.model, Origin, MASE)
#> # A tibble: 8 × 3
#>   .model Origin  MASE
#>   <chr>  <chr>  <dbl>
#> 1 arima  Japan  2.95 
#> 2 arima  NZ     0.576
#> 3 arima  UK     2.31 
#> 4 arima  US     2.46 
#> 5 ets    Japan  2.90 
#> 6 ets    NZ     0.656
#> 7 ets    UK     1.90 
#> 8 ets    US     1.53
fc %>% 
  accuracy(aus_arrivals, by=".model") %>%
  select(.model, MASE)
#> # A tibble: 2 × 2
#>   .model  MASE
#>   <chr>  <dbl>
#> 1 arima   6.49
#> 2 ets     6.38

^{Created on 2022-01-12 by the reprex package (v2.0.1)}

The second table should be the average of the results by model in the first table.