1
$\begingroup$

What is an appropriate way to calculate the MASE accuracy measure (Link) for a time series with multiple seasonalities?

For example: daily data with a strong weekly pattern and annual pattern.

enter image description here

enter image description here

The naive seasonal forecast used to scale the error is only allowing one seasonal pattern m.

Using m = 365: we are not considering the weekly pattern anymore (365 is not a multiple of 7). Using m = 7: we are not considering the annual pattern anymore.

m = 364 considers the weekly pattern and most of the annual pattern. Is that thought right?

Any other idea how to calculate the MASE for multiple seasonalities?

$\endgroup$
1
$\begingroup$

I ended up using m=7 based on a recommendation by Rob J Hyndman in the comments of this post link.

I suggest you pick the shortest of the seasonal periods and use it with a seasonal naive scaling factor.

Thanks again Rob for your advice!

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.