MASE about tsrepr HOT 3 CLOSED

It's alright, thank you for comment. I will make exception (stopping criterion) for lengths of real, forecast and naive arguments of mase function.

from tsrepr.

Good comment, but what to do when N = m, or m > N? For forecasts of seasonal time series, it is not defined really correctly.
I think, the best will be to change the third argument of the mase() function to bench_error, for directly passing an error from naive forecast (or similar).

from tsrepr.

Hmm you are right about this - Hyndman and Koehler do not elaborate very much about seasonality in their paper. But I had a second look at it and saw in their sum-function regarding the naive forcast that they started with i = 2 and, hence, had to set N - 1 (page. 685).
The other formula using "N-m" is around the internet, but not mentioned by them...

Since your calculation of the MASE takes a naive-time-series which is equal in length to the predicted time series, I think your calculation should be right after all. Sorry if I was a bit confused there!

from tsrepr.